[{"data":1,"prerenderedAt":2024},["ShallowReactive",2],{"blog-posts":3},[4,478,786,1031,1379,1671],{"id":5,"title":6,"author":7,"body":8,"date":459,"description":460,"extension":461,"featured":462,"logId":463,"meta":464,"navigation":360,"path":467,"published":360,"readingTime":468,"sector":469,"seo":470,"stem":471,"tags":472,"__hash__":477},"blog\u002Fblog\u002Fcontext-churn-and-the-sisyphus-loop.md","Context Churn & The Loop of Sisyphus: Auditing AI Agent Thrash","RetroAgile Systems Architecture",{"type":9,"value":10,"toc":450},"minimark",[11,15,19,22,30,33,70,80,83,88,91,96,99,102,106,109,113,122,126,133,154,396,399,403,406,443,446],[12,13,6],"h1",{"id":14},"context-churn-the-loop-of-sisyphus-auditing-ai-agent-thrash",[16,17,18],"p",{},"In ancient Greek mythology, King Sisyphus rolled a colossal boulder up a hill, only to watch it roll back down.",[16,20,21],{},"In modern agentic software development, engineering teams pay frontier LLMs to repeat that cycle hundreds of times per sprint.",[16,23,24,25,29],{},"We call this phenomenon ",[26,27,28],"strong",{},"Context Churn",".",[16,31,32],{},"It occurs when an autonomous coding agent, confronted with contradictory requirements, unstated dependencies, or hidden side-effects, enters an oscillating loop of self-correction:",[34,35,36,44,50,59,67],"ol",{},[37,38,39,40,29],"li",{},"Agent implements Feature A in ",[41,42,43],"code",{},"auth.ts",[37,45,46,47,29],{},"Existing unit test fails in ",[41,48,49],{},"billing.ts",[37,51,52,53,55,56,29],{},"Agent modifies ",[41,54,49],{}," to satisfy Feature A, inadvertently breaking ",[41,57,58],{},"tenant.ts",[37,60,61,62,64,65,29],{},"Agent attempts to fix ",[41,63,58],{},", forgets initial constraints from Step 1, and reverts ",[41,66,43],{},[37,68,69],{},"The cycle repeats until context limits trigger compaction or maximum tool calls exhaust the budget.",[71,72,78],"pre",{"className":73,"code":75,"language":76,"meta":77},[74],"language-text","[TELEMETRY STREAM \u002F\u002F LOOP OSCILLATION DETECTED]\nFile: src\u002Fservices\u002Fauth.ts (Modified: 14 times across 22 turns)\nFile: src\u002Fservices\u002Fbilling.ts (Modified: 11 times across 22 turns)\nNet Code Delta: +18 lines (Gross code churn: 1,420 lines modified \u002F reverted)\nTotal Token Ingestion: 340,000 tokens\nCost Incurred: $10.20 for an 18-line diff\nVerdict: REPETITIVE CONTEXT THRASHING\n","text","",[41,79,75],{"__ignoreMap":77},[16,81,82],{},"To traditional project management tools, this sprint task looked active and healthy: commits appeared every four minutes. In reality, the agent was stuck in a feedback loop.",[84,85,87],"h2",{"id":86},"the-anatomy-of-agent-thrash","The Anatomy of Agent Thrash",[16,89,90],{},"Why do autonomous agents thrash? Through telemetry audits across hundreds of repositories, agent churn traces back to three structural failures:",[92,93,95],"h3",{"id":94},"_1-inverted-feedback-signals","1. Inverted Feedback Signals",[16,97,98],{},"When human developers get stuck, they step back, ask a colleague for clarification, or challenge the ticket requirements.",[16,100,101],{},"An LLM agent cannot challenge ambiguous human intent unless explicitly instructed to stop. Instead, it assumes the specification is absolute ground truth and the test harness is broken. It begins mutating surrounding codebase infrastructure: mocking away failing tests, creating redundant wrappers, or silently disabling type checkers.",[92,103,105],{"id":104},"_2-cascading-context-compaction","2. Cascading Context Compaction",[16,107,108],{},"As the conversation transcript grows beyond 60,000 tokens, system prompts and original acceptance criteria suffer from attention decay. The agent prioritizes the most recent compiler error message over the original architectural design.",[92,110,112],{"id":111},"_3-missing-structural-guardrails","3. Missing Structural Guardrails",[16,114,115,116,121],{},"Without deterministic linters and immutable database contracts, the agent treats your entire codebase as malleable clay. Providing an ",[117,118,120],"a",{"href":119},"\u002Fblog\u002Finternal-developer-platform-for-coding-agents","agent-ready internal developer platform"," with isolated database branches prevents speculative migrations from polluting shared environments.",[84,123,125],{"id":124},"auditing-churn-with-continuous-telemetry","Auditing Churn with Continuous Telemetry",[16,127,128,129,132],{},"RetroAgile replaces passive burndowns with real-time ",[26,130,131],{},"Agent Churn Indexes",". By streaming Git commits and AST diffs directly into our telemetry pipeline, we track:",[134,135,136,142,148],"ul",{},[37,137,138,141],{},[26,139,140],{},"Touch Frequency \u002F File Ratio",": A healthy agent feature branch touches a file 1.2 to 2.1 times. Any ratio above 4.0 indicates circular thrashing.",[37,143,144,147],{},[26,145,146],{},"Reversion Velocity",": The rate at which an agent undoes its own previous tool calls within the same branch.",[37,149,150,153],{},[26,151,152],{},"Review Turnaround Latency",": Measuring how long human architects take to approve modular sub-PRs versus monolithic diffs.",[71,155,159],{"className":156,"code":157,"language":158,"meta":77,"style":77},"language-typescript shiki shiki-themes github-dark-default","\u002F\u002F Telemetry Churn Formula\nexport function calculateChurnIndex(branchEvents: GitTelemetryEvent[]): number {\n  const touchesByFile = groupTouches(branchEvents);\n  const reverts = detectReversions(branchEvents);\n  \n  const thrashScore = Object.values(touchesByFile).reduce((acc, count) => {\n    return acc + (count > 2 ? Math.pow(count - 2, 1.8) : 0);\n  }, 0);\n\n  return normalizeScore(thrashScore + reverts.length * 2.5);\n}\n","typescript",[41,160,161,170,210,228,243,249,291,344,355,362,390],{"__ignoreMap":77},[162,163,166],"span",{"class":164,"line":165},"line",1,[162,167,169],{"class":168},"sH3jZ","\u002F\u002F Telemetry Churn Formula\n",[162,171,173,177,180,184,188,192,195,198,201,203,207],{"class":164,"line":172},2,[162,174,176],{"class":175},"suJrU","export",[162,178,179],{"class":175}," function",[162,181,183],{"class":182},"sc3cj"," calculateChurnIndex",[162,185,187],{"class":186},"sZEs4","(",[162,189,191],{"class":190},"sQhOw","branchEvents",[162,193,194],{"class":175},":",[162,196,197],{"class":190}," GitTelemetryEvent",[162,199,200],{"class":186},"[])",[162,202,194],{"class":175},[162,204,206],{"class":205},"sFSAA"," number",[162,208,209],{"class":186}," {\n",[162,211,213,216,219,222,225],{"class":164,"line":212},3,[162,214,215],{"class":175},"  const",[162,217,218],{"class":205}," touchesByFile",[162,220,221],{"class":175}," =",[162,223,224],{"class":182}," groupTouches",[162,226,227],{"class":186},"(branchEvents);\n",[162,229,231,233,236,238,241],{"class":164,"line":230},4,[162,232,215],{"class":175},[162,234,235],{"class":205}," reverts",[162,237,221],{"class":175},[162,239,240],{"class":182}," detectReversions",[162,242,227],{"class":186},[162,244,246],{"class":164,"line":245},5,[162,247,248],{"class":186},"  \n",[162,250,252,254,257,259,262,265,268,271,274,277,280,283,286,289],{"class":164,"line":251},6,[162,253,215],{"class":175},[162,255,256],{"class":205}," thrashScore",[162,258,221],{"class":175},[162,260,261],{"class":186}," Object.",[162,263,264],{"class":182},"values",[162,266,267],{"class":186},"(touchesByFile).",[162,269,270],{"class":182},"reduce",[162,272,273],{"class":186},"((",[162,275,276],{"class":190},"acc",[162,278,279],{"class":186},", ",[162,281,282],{"class":190},"count",[162,284,285],{"class":186},") ",[162,287,288],{"class":175},"=>",[162,290,209],{"class":186},[162,292,294,297,300,303,306,309,312,315,318,321,324,327,329,331,334,336,338,341],{"class":164,"line":293},7,[162,295,296],{"class":175},"    return",[162,298,299],{"class":186}," acc ",[162,301,302],{"class":175},"+",[162,304,305],{"class":186}," (count ",[162,307,308],{"class":175},">",[162,310,311],{"class":205}," 2",[162,313,314],{"class":175}," ?",[162,316,317],{"class":186}," Math.",[162,319,320],{"class":182},"pow",[162,322,323],{"class":186},"(count ",[162,325,326],{"class":175},"-",[162,328,311],{"class":205},[162,330,279],{"class":186},[162,332,333],{"class":205},"1.8",[162,335,285],{"class":186},[162,337,194],{"class":175},[162,339,340],{"class":205}," 0",[162,342,343],{"class":186},");\n",[162,345,347,350,353],{"class":164,"line":346},8,[162,348,349],{"class":186},"  }, ",[162,351,352],{"class":205},"0",[162,354,343],{"class":186},[162,356,358],{"class":164,"line":357},9,[162,359,361],{"emptyLinePlaceholder":360},true,"\n",[162,363,365,368,371,374,376,379,382,385,388],{"class":164,"line":364},10,[162,366,367],{"class":175},"  return",[162,369,370],{"class":182}," normalizeScore",[162,372,373],{"class":186},"(thrashScore ",[162,375,302],{"class":175},[162,377,378],{"class":186}," reverts.",[162,380,381],{"class":205},"length",[162,383,384],{"class":175}," *",[162,386,387],{"class":205}," 2.5",[162,389,343],{"class":186},[162,391,393],{"class":164,"line":392},11,[162,394,395],{"class":186},"}\n",[16,397,398],{},"When an agent's Churn Index crosses the warning threshold, RetroAgile automatically triggers a circuit breaker, halts the autonomous task, and notifies the human architect with exact telemetry on where the specification stalled.",[84,400,402],{"id":401},"the-solution-circuit-breakers-atomic-scopes","The Solution: Circuit Breakers & Atomic Scopes",[16,404,405],{},"How do you eliminate context churn permanently?",[34,407,408,414,437],{},[37,409,410,413],{},[26,411,412],{},"Max Re-touch Quotas",": Restrict agents to a maximum of 3 edits per file per sub-task. If a test is not passing after 3 iterations, the problem is not syntax—it is architecture.",[37,415,416,419,420,426,427,431,432,436],{},[26,417,418],{},"Atomic Contract PRs",": Never task an agent with building an entire subsystem in one pass. Task it with defining the database schema and ",[117,421,425],{"href":422,"rel":423},"https:\u002F\u002Fmihailtd.github.io\u002Fgraphpg\u002F",[424],"nofollow","GraphPG"," subscription interface first, then implement the worker in an isolated sub-PR. Pair contracts with ",[117,428,430],{"href":429},"\u002Fblog\u002Fdeath-of-the-13-point-ticket","spec-readiness scoring"," and ",[117,433,435],{"href":434},"\u002Fblog\u002Fdora-metrics-for-coding-agents","agentic DORA metrics"," to evaluate delivery health.",[37,438,439,442],{},[26,440,441],{},"Continuous Retrospective Audits",": Review your sprint's Churn Index every Friday. If one module consistently induces agent thrash, that module has architectural debt that needs human refactoring.",[16,444,445],{},"Stop paying for LLM circular reasoning. Audit agent telemetry, enforce circuit breakers, and keep your agents moving forward.",[447,448,449],"style",{},"html pre.shiki code .sH3jZ, html code.shiki .sH3jZ{--shiki-default:#8B949E}html pre.shiki code .suJrU, html code.shiki .suJrU{--shiki-default:#FF7B72}html pre.shiki code .sc3cj, html code.shiki .sc3cj{--shiki-default:#D2A8FF}html pre.shiki code .sZEs4, html code.shiki .sZEs4{--shiki-default:#E6EDF3}html pre.shiki code .sQhOw, html code.shiki .sQhOw{--shiki-default:#FFA657}html pre.shiki code .sFSAA, html code.shiki .sFSAA{--shiki-default:#79C0FF}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}",{"title":77,"searchDepth":172,"depth":172,"links":451},[452,457,458],{"id":86,"depth":172,"text":87,"children":453},[454,455,456],{"id":94,"depth":212,"text":95},{"id":104,"depth":212,"text":105},{"id":111,"depth":212,"text":112},{"id":124,"depth":172,"text":125},{"id":401,"depth":172,"text":402},"2026-11-02","Measuring the hidden token and latency costs of ambiguous specifications when agents touch the same files 14 times in circular regression.","md",false,"DISPATCH-002",{"type":465,"canonical":466},"article","https:\u002F\u002Fretroagile.com\u002Fblog\u002Fcontext-churn-and-the-sisyphus-loop","\u002Fblog\u002Fcontext-churn-and-the-sisyphus-loop","6 min","AGENT_TELEMETRY",{"title":6,"description":460},"blog\u002Fcontext-churn-and-the-sisyphus-loop",[473,474,475,476],"agent churn","token burn","telemetry","developer experience","DqUWVYul1XvwUiAO4jSElRvZQhTAdwY90IbbKsRmYCk",{"id":479,"title":480,"author":7,"body":481,"date":771,"description":772,"extension":461,"featured":360,"logId":773,"meta":774,"navigation":360,"path":429,"published":360,"readingTime":776,"sector":777,"seo":778,"stem":779,"tags":780,"__hash__":785},"blog\u002Fblog\u002Fdeath-of-the-13-point-ticket.md","The Death of the 13-Point Ticket: When Coding Agents Move at Light Speed",{"type":9,"value":482,"toc":762},[483,486,489,492,495,498,509,512,518,521,525,528,533,536,542,554,558,562,576,590,594,601,605,612,738,742,745,748,759],[12,484,480],{"id":485},"the-death-of-the-13-point-ticket-when-coding-agents-move-at-light-speed",[16,487,488],{},"For two decades, Scrum sprint planning operated under a single physical constraint: human typing velocity and developer cognitive latency.",[16,490,491],{},"When an engineering team voted on whether a Jira ticket was 3, 5, or 8 story points, they were estimating how long a developer would spend reading legacy code, writing boilerplate, and manually testing edge cases.",[16,493,494],{},"In late 2026, autonomous coding agents severed the link between story points and implementation time.",[16,496,497],{},"An AI coding agent with full workspace access can execute a complex 8-point migration in 180 seconds:",[134,499,500,503,506],{},[37,501,502],{},"Parsing ASTs across 40 files simultaneously.",[37,504,505],{},"Generating types, updating database migration DDL, and writing regression suites in parallel.",[37,507,508],{},"Opening a pull request before the team finishes their morning stand-up.",[16,510,511],{},"Yet that identical agent, handed a supposedly trivial 1-point ticket with ambiguous acceptance criteria, will spend 3 hours spinning in circular context loops. It edits the same file 14 times, burns through API tokens, and produces a regression that halts CI.",[71,513,516],{"className":514,"code":515,"language":76,"meta":77},[74],"[TELEMETRY ANOMALY \u002F\u002F LOG_ID: 984-ALPHA]\nTask: \"Add profile avatar upload, make it fast and secure\"\nPoints Assigned: 1 Story Point (Estimated: 20 min human dev)\nActual Agent Execution:\n- Context Turns: 38 loops\n- Tokens Consumed: 184,200 tokens\n- Files Modified: 12 (7 reverted in thrash)\n- Status: FAILED (Agent assumed local disk storage instead of Cloudflare R2)\n",[41,517,515],{"__ignoreMap":77},[16,519,520],{},"The predictive power of human story points is zero.",[84,522,524],{"id":523},"the-new-currency-spec-readiness-scoring","The New Currency: Spec-Readiness Scoring",[16,526,527],{},"If human typing speed is no longer the bottleneck, what is?",[16,529,530],{},[26,531,532],{},"Specification fidelity.",[16,534,535],{},"In an agentic workflow, velocity depends entirely on the clarity of contract boundaries. If a human systems architect provides an unambiguous specification with explicit input\u002Foutput schemas, database constraints, error codes, and test assertions, autonomous agents execute with high first-pass success.",[71,537,540],{"className":538,"code":539,"language":76,"meta":77},[74],"+-------------------------------------------------------------+\n|                 SPEC-READINESS GATE (PRE-FLIGHT)            |\n+-------------------------------------------------------------+\n| 1. HTTP Route & Contract Interface Defined?         [ PASS ]|\n| 2. PostgreSQL Schema Boundary Explicit?             [ PASS ]|\n| 3. Quantitative Error Codes Catalogued (401, 413)?  [ PASS ]|\n| 4. Ambiguous Adjectives (\"fast\", \"secure\") Banned?  [ PASS ]|\n| 5. Deterministic Acceptance Tests Specified?        [ PASS ]|\n+-------------------------------------------------------------+\n| SPEC READINESS SCORE: 98% -> SAFE FOR AUTONOMOUS DISPATCH   |\n+-------------------------------------------------------------+\n",[41,541,539],{"__ignoreMap":77},[16,543,544,545,549,550,553],{},"When you shift planning from ",[546,547,548],"em",{},"\"How many days will this take a developer?\""," to ",[546,551,552],{},"\"Is this spec verifiable for an autonomous agent?\"",", sprint planning transforms from subjective bargaining into systems engineering.",[84,555,557],{"id":556},"_3-core-rules-for-agentic-planning","3 Core Rules for Agentic Planning",[92,559,561],{"id":560},"_1-ban-ambiguous-adjectives","1. Ban Ambiguous Adjectives",[16,563,564,565,279,568,571,572,575],{},"Adjectives like ",[546,566,567],{},"\"fast\"",[546,569,570],{},"\"scalable\"",", and ",[546,573,574],{},"\"secure\""," cause problems for LLM context windows. They cause probabilistic diffusion—the agent invents defensive caching or arbitrary cryptography that was never requested. Replace adjectives with SLAs:",[134,577,578,584],{},[37,579,580,581],{},"❌ ",[546,582,583],{},"\"Make the endpoint fast.\"",[37,585,586,587],{},"✅ ",[546,588,589],{},"\"p99 latency \u003C 45ms under 500 concurrent requests; cache in Cloudflare KV with 300s TTL.\"",[92,591,593],{"id":592},"_2-lock-schema-first-with-graphpg","2. Lock Schema First with GraphPG",[16,595,596,597,600],{},"Never let an agent guess database columns from ORM abstractions. At RetroAgile, we enforce schema-first development using ",[117,598,425],{"href":422,"rel":599},[424],". The PostgreSQL catalog is introspected natively, creating auto-generated GraphQL SDL and TypeScript types before the feature branch is created. The agent is anchored to a concrete schema contract.",[92,602,604],{"id":603},"_3-acceptance-tests-over-user-stories","3. Acceptance Tests Over User Stories",[16,606,607,608,611],{},"User stories like ",[546,609,610],{},"\"As a user, I want...\""," provide narrative context for humans, but zero verification anchors for autonomous agents. Every ticket must include a deterministic acceptance assertion:",[71,613,615],{"className":156,"code":614,"language":158,"meta":77,"style":77},"\u002F\u002F Deterministic Verification Contract\nit(\"rejects avatar uploads exceeding 5MB with HTTP 413\", async () => {\n  const payload = createMockBuffer(6 * 1024 * 1024);\n  const res = await client.post(\"\u002Fapi\u002Fuser\u002Favatar\", payload);\n  expect(res.status).toBe(413);\n  expect(res.body.error).toBe(\"PAYLOAD_TOO_LARGE\");\n});\n",[41,616,617,622,645,673,699,717,733],{"__ignoreMap":77},[162,618,619],{"class":164,"line":165},[162,620,621],{"class":168},"\u002F\u002F Deterministic Verification Contract\n",[162,623,624,627,629,633,635,638,641,643],{"class":164,"line":172},[162,625,626],{"class":182},"it",[162,628,187],{"class":186},[162,630,632],{"class":631},"s9uIt","\"rejects avatar uploads exceeding 5MB with HTTP 413\"",[162,634,279],{"class":186},[162,636,637],{"class":175},"async",[162,639,640],{"class":186}," () ",[162,642,288],{"class":175},[162,644,209],{"class":186},[162,646,647,649,652,654,657,659,662,664,667,669,671],{"class":164,"line":212},[162,648,215],{"class":175},[162,650,651],{"class":205}," payload",[162,653,221],{"class":175},[162,655,656],{"class":182}," createMockBuffer",[162,658,187],{"class":186},[162,660,661],{"class":205},"6",[162,663,384],{"class":175},[162,665,666],{"class":205}," 1024",[162,668,384],{"class":175},[162,670,666],{"class":205},[162,672,343],{"class":186},[162,674,675,677,680,682,685,688,691,693,696],{"class":164,"line":230},[162,676,215],{"class":175},[162,678,679],{"class":205}," res",[162,681,221],{"class":175},[162,683,684],{"class":175}," await",[162,686,687],{"class":186}," client.",[162,689,690],{"class":182},"post",[162,692,187],{"class":186},[162,694,695],{"class":631},"\"\u002Fapi\u002Fuser\u002Favatar\"",[162,697,698],{"class":186},", payload);\n",[162,700,701,704,707,710,712,715],{"class":164,"line":245},[162,702,703],{"class":182},"  expect",[162,705,706],{"class":186},"(res.status).",[162,708,709],{"class":182},"toBe",[162,711,187],{"class":186},[162,713,714],{"class":205},"413",[162,716,343],{"class":186},[162,718,719,721,724,726,728,731],{"class":164,"line":251},[162,720,703],{"class":182},[162,722,723],{"class":186},"(res.body.error).",[162,725,709],{"class":182},[162,727,187],{"class":186},[162,729,730],{"class":631},"\"PAYLOAD_TOO_LARGE\"",[162,732,343],{"class":186},[162,734,735],{"class":164,"line":293},[162,736,737],{"class":186},"});\n",[84,739,741],{"id":740},"summary-the-architects-role","Summary: The Architect's Role",[16,743,744],{},"Agile is not dead, but the Scrum ritual of planning poker has expired.",[16,746,747],{},"Human software engineers are not typing drones translating Jira tickets into boilerplate. They are systems architects who design specifications, define database boundaries, and review automated telemetry.",[16,749,750,751,754,755,758],{},"Before you dispatch an agent swarm on your codebase, run your tickets through RetroAgile's Spec-Readiness Linter. Pair your contracts with an ",[117,752,753],{"href":119},"internal developer platform built for coding agents"," and guard against circular edits by ",[117,756,757],{"href":467},"tracking context churn",". Stop estimating points. Start engineering contracts.",[447,760,761],{},"html pre.shiki code .sH3jZ, html code.shiki .sH3jZ{--shiki-default:#8B949E}html pre.shiki code .sc3cj, html code.shiki .sc3cj{--shiki-default:#D2A8FF}html pre.shiki code .sZEs4, html code.shiki .sZEs4{--shiki-default:#E6EDF3}html pre.shiki code .s9uIt, html code.shiki .s9uIt{--shiki-default:#A5D6FF}html pre.shiki code .suJrU, html code.shiki .suJrU{--shiki-default:#FF7B72}html pre.shiki code .sFSAA, html code.shiki .sFSAA{--shiki-default:#79C0FF}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}",{"title":77,"searchDepth":172,"depth":172,"links":763},[764,765,770],{"id":523,"depth":172,"text":524},{"id":556,"depth":172,"text":557,"children":766},[767,768,769],{"id":560,"depth":212,"text":561},{"id":592,"depth":212,"text":593},{"id":603,"depth":212,"text":604},{"id":740,"depth":172,"text":741},"2026-11-05","Why human story points fail in autonomous coding swarms, and how Spec-Readiness Scoring prevents multi-thousand dollar context loops.","DISPATCH-001",{"type":465,"canonical":775},"https:\u002F\u002Fretroagile.com\u002Fblog\u002Fdeath-of-the-13-point-ticket","5 min","SPEC_ARCHITECTURE",{"title":480,"description":772},"blog\u002Fdeath-of-the-13-point-ticket",[781,782,783,784],"agile","ai agents","spec-driven planning","scrum reform","6omPwkkOSOKQxgVC_OzagifigUc4LBbiIVd-PP06Td0",{"id":787,"title":788,"author":7,"body":789,"date":1017,"description":1018,"extension":461,"featured":462,"logId":1019,"meta":1020,"navigation":360,"path":434,"published":360,"readingTime":468,"sector":1022,"seo":1023,"stem":1024,"tags":1025,"__hash__":1030},"blog\u002Fblog\u002Fdora-metrics-for-coding-agents.md","DORA Metrics for Autonomous Coding Agents",{"type":9,"value":790,"toc":1008},[791,794,797,800,814,817,820,823,826,832,836,840,843,846,866,872,875,879,882,885,892,903,909,915,919,922,925,931,942,952,956,959,962,976,979,983,986,993,999],[12,792,788],{"id":793},"dora-metrics-for-autonomous-coding-agents",[16,795,796],{},"For nearly a decade, the DevOps Research and Assessment (DORA) framework defined high performance in software delivery.",[16,798,799],{},"Teams tracked four metrics:",[34,801,802,805,808,811],{},[37,803,804],{},"Deployment Frequency",[37,806,807],{},"Lead Time for Changes",[37,809,810],{},"Change Failure Rate",[37,812,813],{},"Time to Restore Service (MTTR)",[16,815,816],{},"In an all-human engineering organization, these four numbers provide clear signal. High deployment frequency reflects small pull request batch sizes. Short lead time indicates efficient continuous integration. Low failure rates and rapid recovery demonstrate stable release automation.",[16,818,819],{},"When autonomous coding agents enter your pipeline, classic DORA metrics lose their diagnostic power.",[16,821,822],{},"An AI agent can author 40 pull requests in an afternoon. If your team measures developer velocity purely by pull request counts and deployment frequency, your dashboards will report record productivity while your human reviewers face a severe review bottleneck.",[16,824,825],{},"To manage agentic software delivery without flying blind, engineering leaders must adapt DORA metrics for autonomous systems.",[71,827,830],{"className":828,"code":829,"language":76,"meta":77},[74],"+-----------------------------------------------------------------------+\n|                    TRADITIONAL DORA vs AGENTIC DORA                   |\n+-----------------------------------------------------------------------+\n| Classic DORA (Human Developers)   | Agentic DORA (Autonomous Agents)  |\n+-----------------------------------+-----------------------------------+\n| Deployment Frequency              | Merged Delivery Velocity          |\n| Lead Time for Changes (Monolith)  | 3-Phase Latency: Spec\u002FAgent\u002FReview|\n| Change Failure Rate               | Agent Change Failure Rate (ACFR)  |\n| Time to Restore Service (MTTR)    | Autonomous Healing vs Escaped MTTR|\n| [Not Tracked]                     | Agent Churn Ratio (ACR)           |\n+-----------------------------------------------------------------------+\n",[41,831,829],{"__ignoreMap":77},[84,833,835],{"id":834},"the-four-agentic-dora-metrics","The Four Agentic DORA Metrics",[92,837,839],{"id":838},"_1-three-phase-lead-time-spec-execution-and-review","1. Three-Phase Lead Time: Spec, Execution, and Review",[16,841,842],{},"Traditional lead time measures the duration from first commit to production deployment. In agentic teams, lumping this timeline into a single number conceals the actual bottleneck.",[16,844,845],{},"RetroAgile separates lead time into three independent stages:",[134,847,848,854,860],{},[37,849,850,853],{},[26,851,852],{},"Specification Latency (Human)",": Time spent clarifying requirements, defining database schemas, and writing acceptance criteria.",[37,855,856,859],{},[26,857,858],{},"Agent Execution Latency (Machine)",": Wall-clock time the agent spends reading context, writing code, and running local tests. In healthy teams, this spans 2 to 15 minutes.",[37,861,862,865],{},[26,863,864],{},"Review Gate Latency (Human)",": Time a completed pull request spends waiting for a human architect's approval.",[71,867,870],{"className":868,"code":869,"language":76,"meta":77},[74],"[SPECIFICATION PHASE]  ───▶  [AGENT RUNTIME]  ───▶  [HUMAN REVIEW GATE]\n   (Human Architect)           (AI Agent)              (Human Senior Eng)\n       2-4 Hours                  3-12 Mins                 18-36 Hours\n",[41,871,869],{"__ignoreMap":77},[16,873,874],{},"When teams track these three intervals separately, the constraint becomes clear: the bottleneck is no longer code authoring. The constraint is prompt ambiguity and human review queue latency.",[92,876,878],{"id":877},"_2-agent-churn-ratio-acr","2. Agent Churn Ratio (ACR)",[16,880,881],{},"Classic DORA has no concept of internal cognitive loops. Humans do not track how many times a developer re-wrote line 42 before committing.",[16,883,884],{},"With autonomous agents, file churn within a single task is the leading indicator of architectural confusion.",[16,886,887,888,891],{},"The ",[26,889,890],{},"Agent Churn Ratio (ACR)"," tracks:",[134,893,894,897,900],{},[37,895,896],{},"Number of execution turns spent on a single task.",[37,898,899],{},"Re-edited files within the same sub-task.",[37,901,902],{},"Reverted code hunks before the pull request opens.",[71,904,907],{"className":905,"code":906,"language":76,"meta":77},[74],"ACR = (Total File Modifications - Net Files Changed) \u002F Net Files Changed\n",[41,908,906],{"__ignoreMap":77},[16,910,911,912,29],{},"Scores below 0.25 reflect direct execution without thrashing. Once the ratio climbs past 1.50, the agent is thrashing—guessing database columns, battling contradictory prompts, or repeating failing test runs. This is the core fingerprint of ",[117,913,914],{"href":467},"context churn and the Sisyphus loop",[92,916,918],{"id":917},"_3-agent-change-failure-rate-acfr","3. Agent Change Failure Rate (ACFR)",[16,920,921],{},"In human workflows, change failure rate measures the percentage of deployments requiring a hotfix or rollback.",[16,923,924],{},"In agentic pipelines, we measure failure earlier: at the automated verification gate.",[16,926,887,927,930],{},[26,928,929],{},"Agent Change Failure Rate (ACFR)"," measures the percentage of agent-generated pull requests that:",[34,932,933,936,939],{},[37,934,935],{},"Fail automated unit, contract, or type checks on first submission.",[37,937,938],{},"Require human intervention to correct architectural mistakes.",[37,940,941],{},"Cause production rollbacks after merge.",[16,943,944,945,431,948,951],{},"When a repository enforces schema-first development through ",[117,946,425],{"href":422,"rel":947},[424],[117,949,950],{"href":429},"Spec-Readiness linting",", first-pass pass rates exceed 92%. When tasks enter with loose requirements, ACFR climbs past 40%, generating human review fatigue.",[92,953,955],{"id":954},"_4-autonomous-healing-mttr-versus-escaped-mttr","4. Autonomous Healing MTTR versus Escaped MTTR",[16,957,958],{},"When an automated build fails, how quickly does the system recover?",[16,960,961],{},"In agentic teams, recovery divides into two categories:",[134,963,964,970],{},[37,965,966,969],{},[26,967,968],{},"Autonomous Self-Healing MTTR",": When CI fails on a feature branch, the agent receives the test trace and applies a fix automatically. Target: under 4 minutes.",[37,971,972,975],{},[26,973,974],{},"Escaped Defect MTTR",": A bug bypasses testing and enters main or production, requiring human triage, root-cause investigation, and emergency patching.",[16,977,978],{},"If your escaped defect rate increases while deployment frequency climbs, your team is generating technical debt faster than humans can review it.",[84,980,982],{"id":981},"implementing-agent-telemetry","Implementing Agent Telemetry",[16,984,985],{},"Tracking these metrics manually on spreadsheets is impossible. Agent workflows generate thousands of fine-grained events across git commits, language model tokens, and continuous integration runs.",[16,987,988,989,992],{},"RetroAgile ingests raw Git webhooks and agent spans into PostgreSQL 18. Using ",[117,990,425],{"href":422,"rel":991},[424],", the platform streams real-time GraphQL change-feeds directly to engineering dashboards. Teams spot agent churn spikes and review bottlenecks immediately rather than waiting for bi-weekly sprint reviews.",[16,994,995,996,998],{},"Agile estimation may be obsolete, but engineering rigor is not. Ground your autonomous coding agents in an ",[117,997,120],{"href":119},", track agentic DORA metrics, and focus human effort where it matters most: system architecture and contract clarity.",[16,1000,1001,1002,1007],{},"For organizations building scalable agent teams and real-time data engines, our partners at ",[117,1003,1006],{"href":1004,"rel":1005},"https:\u002F\u002Finnovi.pro",[424],"Innovi Pro"," provide architectural guidance.",{"title":77,"searchDepth":172,"depth":172,"links":1009},[1010,1016],{"id":834,"depth":172,"text":835,"children":1011},[1012,1013,1014,1015],{"id":838,"depth":212,"text":839},{"id":877,"depth":212,"text":878},{"id":917,"depth":212,"text":918},{"id":954,"depth":212,"text":955},{"id":981,"depth":172,"text":982},"2026-11-07","Traditional DORA metrics assume human developer typing speed is the primary delivery constraint. Here is how engineering teams adapt DORA for autonomous agents.","METRICS-001",{"type":465,"canonical":1021},"https:\u002F\u002Fretroagile.com\u002Fblog\u002Fdora-metrics-for-coding-agents","AGENT_METRICS",{"title":788,"description":1018},"blog\u002Fdora-metrics-for-coding-agents",[1026,1027,1028,1029],"dora metrics","agentic workflows","git telemetry","engineering management","fjXtZOonosB6fsuHWCCJDMm370gwmAhwUecUzw6YWxU",{"id":1032,"title":1033,"author":7,"body":1034,"date":1362,"description":1363,"extension":461,"featured":462,"logId":1364,"meta":1365,"navigation":360,"path":1368,"published":360,"readingTime":468,"sector":1369,"seo":1370,"stem":1371,"tags":1372,"__hash__":1378},"blog\u002Fblog\u002Fephemeral-demo-infrastructure-graphpg-drizzle-cloudflare.md","Ephemeral Zero Trust Demo Infrastructure with GraphPG, Drizzle, and Cloudflare Tunnel",{"type":9,"value":1035,"toc":1357},[1036,1039,1052,1060,1064,1084,1116,1120,1147,1188,1192,1199,1291,1313,1347,1354],[12,1037,1033],{"id":1038},"ephemeral-zero-trust-demo-infrastructure-with-graphpg-drizzle-and-cloudflare-tunnel",[16,1040,1041,1042,1046,1047,1051],{},"Live telemetry sandboxes usually force an uncomfortable choice between fake browser state that never touches a real query planner and dedicated cloud servers that cost money while idle. Standard staging clusters solve the realism gap by leaving managed databases online all month, which raises fixed hosting overhead and forces manual credential rotation or SSH cleanup when sample data goes stale. At ",[117,1043,1045],{"href":1044},"\u002Fdemo","RetroAgile",", we wired the sandbox directly to a disposable PostgreSQL 18 and ",[117,1048,1050],{"href":422,"rel":1049},[424],"GraphPG 1.3.0"," backend defined in code, routed exclusively over Cloudflare Zero Trust Service Tokens, and backed by deterministic Worker fallback when the server is destroyed.",[16,1053,1054],{},[1055,1056],"img",{"alt":1057,"src":1058,"title":1059},"Request path from RetroAgile Nuxt Worker on Cloudflare through Cloudflare Access Service Tokens and an outbound Argo Tunnel into GraphPG 1.3.0 and PostgreSQL 18, with deterministic Worker fallback when the VPS is torn down.","\u002Fblog\u002Fdiagrams\u002Fretroagile-demo-tunnel-poster.webp","Live GraphPG execution when the VPS is up, seamless deterministic fallback when torn down.",[84,1061,1063],{"id":1062},"why-ephemeral-workers-need-an-inside-out-database-path","Why ephemeral Workers need an inside-out database path",[16,1065,1066,1067,1069,1070,1075,1076,1079,1080,1083],{},"Browser visitors interacting with ",[41,1068,1044],{}," should never hold database URLs or origin API keys, and a sandbox VM should never expose PostgreSQL port 5432 or GraphQL endpoints to public internet scanners. Instead, our ",[117,1071,1074],{"href":1072,"rel":1073},"https:\u002F\u002Fgithub.com\u002Fmihailtd\u002Fretroagile\u002Ftree\u002Fmain\u002Fpackages\u002Finfra-demo",[424],"infra-demo IaC package"," provisions an UpCloud ",[41,1077,1078],{},"DEV-1xCPU-1GB-10GB"," server with ",[41,1081,1082],{},"default_action = \"drop\""," firewall rules for both IPv4 and IPv6, blocking every inbound socket.",[16,1085,1086,1087,1090,1091,549,1094,1097,1098,1101,1102,431,1105,1108,1109,431,1112,1115],{},"Inside the VM, ",[41,1088,1089],{},"cloudflared"," maintains an outbound tunnel to Cloudflare Edge and routes ",[41,1092,1093],{},"demo-api.retroagile.com",[41,1095,1096],{},"http:\u002F\u002F127.0.0.1:8081",". Cloudflare Access sits in front of that hostname with a ",[41,1099,1100],{},"nonIdentity"," policy requiring ",[41,1103,1104],{},"CF-Access-Client-Id",[41,1106,1107],{},"CF-Access-Client-Secret"," headers injected only inside RetroAgile's server routes (",[41,1110,1111],{},"\u002Fapi\u002Fdemo\u002Fgraphpg",[41,1113,1114],{},"\u002Fapi\u002Fdemo\u002Freset",").",[84,1117,1119],{"id":1118},"co-hosting-retroagile-and-innovi-care-on-one-postgres-instance","Co-hosting RetroAgile and Innovi Care on one Postgres instance",[16,1121,1122,1123,1126,1127,1130,1131,1134,1135,1138,1139,1142,1143,1146],{},"Running separate VMs or managed instances for multiple demo portals wastes memory before a visitor opens either page. Because PostgreSQL isolates catalogs and memory pools cleanly across logical databases, one ",[41,1124,1125],{},"postgres:18-alpine"," container (",[41,1128,1129],{},"320m"," memory ceiling) hosts both ",[41,1132,1133],{},"retroagile"," telemetry tables and ",[41,1136,1137],{},"innovi_care"," FHIR R4 tables side by side, each served by its own ",[41,1140,1141],{},"graphpg"," 1.3.0 binary (",[41,1144,1145],{},"96m"," ceiling each).",[16,1148,1149,1150,1155,1156,1161,1162,1165,1166,279,1169,279,1172,279,1175,279,1178,571,1181,1184,1185,29],{},"You can read the container memory budgeting breakdown on ",[117,1151,1154],{"href":1152,"rel":1153},"https:\u002F\u002Fmihai.ltd\u002Fblog\u002Frunning-two-graphpg-apis-on-a-3-dollar-vps",[424],"mihai.ltd"," and the broader clinical security model in ",[117,1157,1160],{"href":1158,"rel":1159},"https:\u002F\u002Finnovi.pro\u002Fblog\u002Fzero-trust-demo-environments-cloudflare-upcloud-graphpg",[424],"the Innovi Pro zero-trust blueprint",". On RetroAgile's side, ",[41,1163,1164],{},"graphpg.yaml"," binds the telemetry catalog (",[41,1167,1168],{},"repositories",[41,1170,1171],{},"integration_connections",[41,1173,1174],{},"spec_audits",[41,1176,1177],{},"agent_sessions",[41,1179,1180],{},"agent_telemetry_events",[41,1182,1183],{},"pull_request_lifecycles",") plus the volatile mutation ",[41,1186,1187],{},"resetRetroagileDemo",[84,1189,1191],{"id":1190},"deterministic-drizzle-migrations-and-single-sql-inspection","Deterministic Drizzle migrations and single-SQL inspection",[16,1193,1194,1195,1198],{},"Rather than configuring snapshots or replication slots on an ephemeral server, both schema creation and clean resets live in Drizzle SQL migrations (",[41,1196,1197],{},"packages\u002Fdb\u002Fmigrations\u002F0001_demo_reset_rpc.sql","):",[71,1200,1204],{"className":1201,"code":1202,"language":1203,"meta":77,"style":77},"language-sql shiki shiki-themes github-dark-default","CREATE OR REPLACE FUNCTION reset_retroagile_demo()\nRETURNS jsonb\nLANGUAGE plpgsql\nVOLATILE\nAS $$\nDECLARE\n  v_now timestamptz := clock_timestamp();\nBEGIN\n  TRUNCATE TABLE\n    agent_telemetry_events, pull_request_lifecycles,\n    agent_sessions, spec_audits, integration_connections, repositories\n  RESTART IDENTITY CASCADE;\n  -- deterministic seed rows follow using v_now intervals\n  RETURN jsonb_build_object('status', 'reset', 'database', 'retroagile', 'reset_at', v_now);\nEND;\n$$;\n","sql",[41,1205,1206,1211,1216,1221,1226,1231,1236,1241,1246,1251,1256,1261,1267,1273,1279,1285],{"__ignoreMap":77},[162,1207,1208],{"class":164,"line":165},[162,1209,1210],{},"CREATE OR REPLACE FUNCTION reset_retroagile_demo()\n",[162,1212,1213],{"class":164,"line":172},[162,1214,1215],{},"RETURNS jsonb\n",[162,1217,1218],{"class":164,"line":212},[162,1219,1220],{},"LANGUAGE plpgsql\n",[162,1222,1223],{"class":164,"line":230},[162,1224,1225],{},"VOLATILE\n",[162,1227,1228],{"class":164,"line":245},[162,1229,1230],{},"AS $$\n",[162,1232,1233],{"class":164,"line":251},[162,1234,1235],{},"DECLARE\n",[162,1237,1238],{"class":164,"line":293},[162,1239,1240],{},"  v_now timestamptz := clock_timestamp();\n",[162,1242,1243],{"class":164,"line":346},[162,1244,1245],{},"BEGIN\n",[162,1247,1248],{"class":164,"line":357},[162,1249,1250],{},"  TRUNCATE TABLE\n",[162,1252,1253],{"class":164,"line":364},[162,1254,1255],{},"    agent_telemetry_events, pull_request_lifecycles,\n",[162,1257,1258],{"class":164,"line":392},[162,1259,1260],{},"    agent_sessions, spec_audits, integration_connections, repositories\n",[162,1262,1264],{"class":164,"line":1263},12,[162,1265,1266],{},"  RESTART IDENTITY CASCADE;\n",[162,1268,1270],{"class":164,"line":1269},13,[162,1271,1272],{},"  -- deterministic seed rows follow using v_now intervals\n",[162,1274,1276],{"class":164,"line":1275},14,[162,1277,1278],{},"  RETURN jsonb_build_object('status', 'reset', 'database', 'retroagile', 'reset_at', v_now);\n",[162,1280,1282],{"class":164,"line":1281},15,[162,1283,1284],{},"END;\n",[162,1286,1288],{"class":164,"line":1287},16,[162,1289,1290],{},"$$;\n",[16,1292,1293,1294,1297,1298,1301,1302,1305,1306,1309,1310,1312],{},"PRO Tip #1: Placing the ",[41,1295,1296],{},"TRUNCATE ... RESTART IDENTITY CASCADE"," and seed inserts inside one ",[41,1299,1300],{},"VOLATILE"," PL\u002FpgSQL function ensures that the Docker boot initializer, the 30-minute host cron script, the local ",[41,1303,1304],{},"pnpm run db:seed"," CLI, and the ",[26,1307,1308],{},"Reset Demo DB"," button on ",[41,1311,1044],{}," execute identical SQL logic.",[16,1314,1315,1316,1319,1320,1322,1323,1326,1327,1330,1331,1334,1335,1338,1339,1342,1343,1346],{},"When visitors click ",[26,1317,1318],{},"Query GraphPG Tunnel"," on ",[41,1321,1044],{},", the server endpoint checks ",[41,1324,1325],{},"GRAPHPG_URL",". When the UpCloud server is provisioned, GraphPG executes the GraphQL selection in a single lateral PostgreSQL 18 statement (",[41,1328,1329],{},"mode: \"live-vps\"","). When ",[41,1332,1333],{},"pulumi destroy"," or ",[41,1336,1337],{},"tofu destroy"," tears down the server between demos, the endpoint automatically degrades to ",[41,1340,1341],{},"mode: \"deterministic-fallback\"",", returning the same seeded dataset, ",[41,1344,1345],{},"reset_retroagile_demo()"," JSONB contract, and compiled SQL receipt with zero broken pages. Note that live UpCloud provisioning itself has not been executed yet in this repo so compute spend stays at zero until explicitly brought up.",[16,1348,1349,1350,29],{},"What else can you build with this setup? Swap in Neon or Fly.io Postgres if you prefer serverless scaling over fixed VPS compute, or wire your own CI preview branches to tear down demo backends nightly while keeping edge docs online. For architectural assessments of agentic developer platforms or database-first APIs, reach out through ",[117,1351,1006],{"href":1352,"rel":1353},"https:\u002F\u002Finnovi.pro\u002F#contact",[424],[447,1355,1356],{},"html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}",{"title":77,"searchDepth":172,"depth":172,"links":1358},[1359,1360,1361],{"id":1062,"depth":172,"text":1063},{"id":1118,"depth":172,"text":1119},{"id":1190,"depth":172,"text":1191},"2026-11-14","How RetroAgile pairs Cloudflare Workers with a disposable PostgreSQL 18 and GraphPG 1.3.0 backend on UpCloud using Drizzle SQL reset RPCs and Cloudflare Access Service Tokens.","DISPATCH-005",{"type":465,"image":1058,"imageFallback":1366,"canonical":1367},"\u002Fblog\u002Fdiagrams\u002Fretroagile-demo-tunnel-poster.png","https:\u002F\u002Fretroagile.com\u002Fblog\u002Fephemeral-demo-infrastructure-graphpg-drizzle-cloudflare","\u002Fblog\u002Fephemeral-demo-infrastructure-graphpg-drizzle-cloudflare","INFRASTRUCTURE",{"title":1033,"description":1363},"blog\u002Fephemeral-demo-infrastructure-graphpg-drizzle-cloudflare",[1141,1373,1374,1375,1376,1377],"drizzle orm","cloudflare tunnel","upcloud","postgresql","zero trust","h0ktsn9Jh4x4lLomM9FrdBHG0nHhLNbQMMdq5wRu85I",{"id":1380,"title":1381,"author":1382,"body":1383,"date":1657,"description":1658,"extension":461,"featured":462,"logId":1659,"meta":1660,"navigation":360,"path":1662,"published":360,"readingTime":1663,"sector":1664,"seo":1665,"stem":1666,"tags":1667,"__hash__":1670},"blog\u002Fblog\u002Finstant-telemetry-with-postgresql-and-graphpg.md","Architecting Continuous Retrospectives: Low-Latency Telemetry with GraphPG","RetroAgile Core Team",{"type":9,"value":1384,"toc":1647},[1385,1388,1391,1394,1397,1400,1403,1414,1421,1425,1428,1434,1438,1442,1449,1456,1519,1527,1530,1547,1572,1575,1579,1589,1599,1603,1606,1609,1631,1634,1645],[12,1386,1381],{"id":1387},"architecting-continuous-retrospectives-low-latency-telemetry-with-graphpg",[16,1389,1390],{},"Developer productivity dashboards often lag behind real-time engineering work.",[16,1392,1393],{},"Many platforms poll the GitHub REST API on 15-minute intervals, write unstructured JSON blobs into document stores, run batch jobs overnight, and display yesterday's velocity metrics on a delayed dashboard.",[16,1395,1396],{},"When a team of 4 human engineers coordinates 15 autonomous AI agents, that batch model breaks down.",[16,1398,1399],{},"In an active sprint, agents push commits every 90 seconds, run parallel CI checks across 8 branches, and generate hundreds of telemetry events per hour. If your retrospective platform takes 20 minutes to notice a CI failure cluster or agent churn storm, the damage—wasted tokens, blocked PR queues, and broken main branches—has already occurred.",[16,1401,1402],{},"RetroAgile required an architecture with three concrete guarantees:",[34,1404,1405,1408,1411],{},[37,1406,1407],{},"Sub-second event ingestion: Immediate reaction to Git pushes, PR reviews, and CI runs.",[37,1409,1410],{},"Zero ORM bloat: Direct relational mapping with zero query-compilation overhead.",[37,1412,1413],{},"Native Real-Time Subscriptions: Instant change-feeds pushed directly to browser dashboards and Slack bots.",[16,1415,1416,1417,1420],{},"Here is why we built RetroAgile's telemetry engine on ",[117,1418,425],{"href":422,"rel":1419},[424]," and PostgreSQL 18.",[84,1422,1424],{"id":1423},"the-native-postgresql-advantage","The Native PostgreSQL Advantage",[16,1426,1427],{},"Rather than maintaining a pipeline of Kafka, Redis pub\u002Fsub, MongoDB, and custom GraphQL resolvers, RetroAgile runs on native PostgreSQL schemas introspected directly by GraphPG.",[71,1429,1432],{"className":1430,"code":1431,"language":76,"meta":77},[74],"[GIT \u002F CI TELEMETRY EVENT STREAM]\n               │\n               ▼ (Webhook \u002F CLI Pre-Flight)\n     [Cloudflare Worker Gateway]\n               │\n               ▼ (TLS Connection Pooling)\n   [PostgreSQL 18 Telemetry Hub]\n      • Native UUIDv7 for non-bloating time-ordered PKs\n      • TSTZRANGE for PR lifecycle and review durations\n      • JSON_TABLE virtual views for multi-valued child events\n               │\n               ▼ (Native Catalog Introspection)\n      [GraphPG Engine Core]\n      • Instant GraphQL SDL generation via pg_type & information_schema\n      • Low-latency subscriptions via PostgreSQL LISTEN\u002FNOTIFY & WAL streaming\n               │\n               ▼ (WebSocket \u002F Server-Sent Events)\n [RetroAgile Dashboard & Autonomous Synthesis Agents]\n",[41,1433,1431],{"__ignoreMap":77},[84,1435,1437],{"id":1436},"_3-architectural-pillars-of-the-retroagile-telemetry-stack","3 Architectural Pillars of the RetroAgile Telemetry Stack",[92,1439,1441],{"id":1440},"_1-postgresql-18-native-uuidv7-primary-keys","1. PostgreSQL 18 Native UUIDv7 Primary Keys",[16,1443,1444,1445,1448],{},"Telemetry events are high-volume, strictly time-ordered, and append-heavy. Traditional ",[41,1446,1447],{},"uuidv4()"," primary keys cause B-tree page fragmentation and index bloat over millions of events.",[16,1450,1451,1452,1455],{},"By standardizing on native ",[41,1453,1454],{},"uuidv7()",", every telemetry record is naturally ordered by wall-clock timestamp while preserving globally unique entropy across distributed agent workers. B-tree insertions append cleanly to the right side of the index with zero page splits.",[71,1457,1459],{"className":1201,"code":1458,"language":1203,"meta":77,"style":77},"CREATE TABLE agent_telemetry_events (\n    id UUID PRIMARY KEY DEFAULT uuidv7(),\n    repository_id UUID NOT NULL,\n    agent_session_id UUID NOT NULL,\n    event_type VARCHAR(64) NOT NULL,\n    token_usage_input INTEGER NOT NULL DEFAULT 0,\n    token_usage_output INTEGER NOT NULL DEFAULT 0,\n    recorded_at TIMESTAMPTZ NOT NULL DEFAULT clock_timestamp()\n);\n\nCREATE INDEX idx_telemetry_repo_time \nON agent_telemetry_events (repository_id, recorded_at DESC);\n",[41,1460,1461,1466,1471,1476,1481,1486,1491,1496,1501,1505,1509,1514],{"__ignoreMap":77},[162,1462,1463],{"class":164,"line":165},[162,1464,1465],{},"CREATE TABLE agent_telemetry_events (\n",[162,1467,1468],{"class":164,"line":172},[162,1469,1470],{},"    id UUID PRIMARY KEY DEFAULT uuidv7(),\n",[162,1472,1473],{"class":164,"line":212},[162,1474,1475],{},"    repository_id UUID NOT NULL,\n",[162,1477,1478],{"class":164,"line":230},[162,1479,1480],{},"    agent_session_id UUID NOT NULL,\n",[162,1482,1483],{"class":164,"line":245},[162,1484,1485],{},"    event_type VARCHAR(64) NOT NULL,\n",[162,1487,1488],{"class":164,"line":251},[162,1489,1490],{},"    token_usage_input INTEGER NOT NULL DEFAULT 0,\n",[162,1492,1493],{"class":164,"line":293},[162,1494,1495],{},"    token_usage_output INTEGER NOT NULL DEFAULT 0,\n",[162,1497,1498],{"class":164,"line":346},[162,1499,1500],{},"    recorded_at TIMESTAMPTZ NOT NULL DEFAULT clock_timestamp()\n",[162,1502,1503],{"class":164,"line":357},[162,1504,343],{},[162,1506,1507],{"class":164,"line":364},[162,1508,361],{"emptyLinePlaceholder":360},[162,1510,1511],{"class":164,"line":392},[162,1512,1513],{},"CREATE INDEX idx_telemetry_repo_time \n",[162,1515,1516],{"class":164,"line":1263},[162,1517,1518],{},"ON agent_telemetry_events (repository_id, recorded_at DESC);\n",[92,1520,1522,1523,1526],{"id":1521},"_2-lifecycles-as-native-range-types-tstzrange","2. Lifecycles as Native Range Types (",[41,1524,1525],{},"TSTZRANGE",")",[16,1528,1529],{},"Measuring PR turnaround latency and review bottlenecks requires queries with complex boundary checks across separate timestamps.",[16,1531,1532,1533,1535,1536,1539,1540,1543,1544,1198],{},"RetroAgile maps PR active lifecycles and agent tool execution intervals directly to PostgreSQL ",[41,1534,1525],{}," (",[41,1537,1538],{},"tstzrange","). This allows GraphPG to auto-generate ",[41,1541,1542],{},"DatetimeRange"," GraphQL filters and gives our query engine native access to the GiST overlap operator (",[41,1545,1546],{},"&&",[71,1548,1550],{"className":1201,"code":1549,"language":1203,"meta":77,"style":77},"-- Find all agent sessions that overlapped with the CI outage window\nSELECT agent_session_id, execution_window\nFROM agent_execution_spans\nWHERE execution_window && tstzrange('2026-11-07 08:00:00Z', '2026-11-07 09:30:00Z');\n",[41,1551,1552,1557,1562,1567],{"__ignoreMap":77},[162,1553,1554],{"class":164,"line":165},[162,1555,1556],{},"-- Find all agent sessions that overlapped with the CI outage window\n",[162,1558,1559],{"class":164,"line":172},[162,1560,1561],{},"SELECT agent_session_id, execution_window\n",[162,1563,1564],{"class":164,"line":212},[162,1565,1566],{},"FROM agent_execution_spans\n",[162,1568,1569],{"class":164,"line":230},[162,1570,1571],{},"WHERE execution_window && tstzrange('2026-11-07 08:00:00Z', '2026-11-07 09:30:00Z');\n",[16,1573,1574],{},"GiST index lookups resolve overlaps without table scans or slow date arithmetic.",[92,1576,1578],{"id":1577},"_3-zero-glue-graphql-with-graphpg","3. Zero-Glue GraphQL with GraphPG",[16,1580,1581,1582,279,1585,1588],{},"Instead of writing and maintaining hundreds of hand-coded GraphQL schema definitions, resolvers, and dataloaders, GraphPG inspects the PostgreSQL system catalogs (",[41,1583,1584],{},"pg_type",[41,1586,1587],{},"information_schema.columns",") and auto-generates the GraphQL API dynamically.",[16,1590,1591,1592,1334,1595,1598],{},"When database engineers add an ",[41,1593,1594],{},"agent_churn_score",[41,1596,1597],{},"token_cost_usd"," column via a migration, GraphPG discovers the type alteration instantly. The frontend dashboard receives typed GraphQL queries and real-time subscription feeds without touching a single backend resolver.",[84,1600,1602],{"id":1601},"real-time-retrospectives-in-action","Real-Time Retrospectives in Action",[16,1604,1605],{},"Because PostgreSQL change-feeds stream continuously into GraphPG, RetroAgile's synthesis agent stays perpetually aware of the sprint's trajectory.",[16,1607,1608],{},"When a pull request merges:",[34,1610,1611,1614,1617,1620],{},[37,1612,1613],{},"The GitHub webhook hits the Cloudflare Worker.",[37,1615,1616],{},"The Worker inserts the telemetry span into PostgreSQL.",[37,1618,1619],{},"GraphPG dispatches a real-time GraphQL subscription event.",[37,1621,1622,1623,1626,1627,1630],{},"The RetroAgile synthesis agent updates the sprint's review bottleneck distribution and churn matrix in under 400 milliseconds. Teams monitor these streams to evaluate ",[117,1624,1625],{"href":434},"DORA metrics for coding agents"," and flag ",[117,1628,1629],{"href":467},"context churn"," before circular loops escalate.",[16,1632,1633],{},"This is the benefit of combining a relational kernel with GraphQL auto-introspection.",[16,1635,1636,1637,1640,1641,1644],{},"For teams looking to build low-latency data pipelines without ORM compromises, ",[117,1638,425],{"href":422,"rel":1639},[424]," provides instant GraphQL introspection directly on PostgreSQL. For organizations designing mission-critical enterprise systems, our advisory team at ",[117,1642,1006],{"href":1004,"rel":1643},[424]," provides hands-on architectural advisory.",[447,1646,1356],{},{"title":77,"searchDepth":172,"depth":172,"links":1648},[1649,1650,1656],{"id":1423,"depth":172,"text":1424},{"id":1436,"depth":172,"text":1437,"children":1651},[1652,1653,1655],{"id":1440,"depth":212,"text":1441},{"id":1521,"depth":212,"text":1654},"2. Lifecycles as Native Range Types (TSTZRANGE)",{"id":1577,"depth":212,"text":1578},{"id":1601,"depth":172,"text":1602},"2026-10-28","How RetroAgile leverages native PostgreSQL catalogs and GraphPG real-time change feeds to synthesize sprint retrospectives without polling overhead.","DISPATCH-003",{"type":465,"canonical":1661},"https:\u002F\u002Fretroagile.com\u002Fblog\u002Finstant-telemetry-with-postgresql-and-graphpg","\u002Fblog\u002Finstant-telemetry-with-postgresql-and-graphpg","7 min","DATABASE_KERNEL",{"title":1381,"description":1658},"blog\u002Finstant-telemetry-with-postgresql-and-graphpg",[1376,1141,1668,1669],"real-time telemetry","software architecture","U2siRGIdrjWp2vHBTSFi9OZWMeKwEPtmGo38j_aj4tA",{"id":1672,"title":1673,"author":7,"body":1674,"date":2011,"description":2012,"extension":461,"featured":462,"logId":2013,"meta":2014,"navigation":360,"path":119,"published":360,"readingTime":468,"sector":2016,"seo":2017,"stem":2018,"tags":2019,"__hash__":2023},"blog\u002Fblog\u002Finternal-developer-platform-for-coding-agents.md","Building an Internal Developer Platform for Coding Agents",{"type":9,"value":1675,"toc":2002},[1676,1679,1682,1685,1688,1691,1694,1700,1704,1707,1711,1714,1717,1720,1731,1735,1741,1747,1755,1766,1775,1900,1904,1907,1910,1913,1927,1936,1940,1943,1946,1949,1955,1964,1968,1971,1986,1999],[12,1677,1673],{"id":1678},"building-an-internal-developer-platform-for-coding-agents",[16,1680,1681],{},"For the past five years, enterprise platform engineering focused on a single customer: the human software engineer.",[16,1683,1684],{},"Organizations spent millions deploying internal developer platforms (IDPs) centered around Spotify Backstage, web service catalogs, and interactive scaffolding wizards. These platforms solved real human problems: discovering documentation, requesting cloud credentials, and standardizing Kubernetes manifests through point-and-click interfaces.",[16,1686,1687],{},"Autonomous coding agents break every core assumption of the human IDP.",[16,1689,1690],{},"An AI agent does not browse a service catalog portal, read unversioned wiki pages, or fill out multi-step web forms. When an agent receives a coding assignment, it requires machine-readable contract specifications, ephemeral execution environments, and pre-authenticated tool registries.",[16,1692,1693],{},"If your platform engineering team fails to adapt its IDP for agentic consumers, your autonomous developers will stall, burn API tokens, or introduce regressions into shared staging infrastructure.",[71,1695,1698],{"className":1696,"code":1697,"language":76,"meta":77},[74],"+-----------------------------------------------------------------------+\n|                    TRADITIONAL IDP vs AGENTIC IDP                     |\n+-----------------------------------------------------------------------+\n| Human IDP (Backstage \u002F Portals)   | Agentic IDP (RetroAgile \u002F MCP)    |\n+-----------------------------------+-----------------------------------+\n| Visual web dashboards & forms     | Structured JSON\u002FYAML schema APIs  |\n| Markdown wikis and Confluence     | Native DB catalog introspection   |\n| Shared long-lived staging envs    | Ephemeral, branch-isolated DBs    |\n| Manual code review checklists     | Automated contract verification   |\n| Unbounded developer sessions      | Explicit context & turn quotas    |\n+-----------------------------------------------------------------------+\n",[41,1699,1697],{"__ignoreMap":77},[84,1701,1703],{"id":1702},"the-four-foundations-of-an-agent-ready-idp","The Four Foundations of an Agent-Ready IDP",[16,1705,1706],{},"Teams moving past visual dashboards often ask what an agent-ready internal developer platform looks like and how coding agents should interact with an agentic internal developer portal. Instead of browsing web forms, an AI coding agent operates through four deterministic primitives.",[92,1708,1710],{"id":1709},"_1-ephemeral-copy-on-write-database-environments","1. Ephemeral, Copy-on-Write Database Environments",[16,1712,1713],{},"A human engineer can work productively on a shared development database, coordinating table migrations over Slack.",[16,1715,1716],{},"An autonomous coding agent cannot. If an agent executes a speculative database migration on a shared database instance, it risks corrupting schema state for every other engineer and automated test runner.",[16,1718,1719],{},"An agent-ready IDP must spin up an isolated database branch for every agent task in under five seconds. When the agent starts a task:",[34,1721,1722,1725,1728],{},[37,1723,1724],{},"The platform provisions an isolated PostgreSQL worktree or copy-on-write branch.",[37,1726,1727],{},"The agent executes DDL migrations, seeds synthetic fixtures, and validates queries locally.",[37,1729,1730],{},"If the agent fails its verification gate, the branch is discarded immediately with zero state pollution.",[92,1732,1734],{"id":1733},"_2-schema-first-catalog-introspection-with-graphpg","2. Schema-First Catalog Introspection with GraphPG",[16,1736,1737,1738,29],{},"Agents perform poorly when forced to guess database schemas from fragmented ORM models or outdated markdown documentation. Incomplete schema context leads directly to ",[117,1739,1740],{"href":467},"context churn and circular loops",[16,1742,1743,1744,29],{},"Rather than feeding loose documentation into an LLM context window, an agent IDP exposes native schema catalogs directly. RetroAgile enforces schema-first contract delivery using ",[117,1745,425],{"href":422,"rel":1746},[424],[16,1748,1749,1750,279,1752,1754],{},"GraphPG inspects PostgreSQL system catalogs (",[41,1751,1584],{},[41,1753,1587],{},") and exposes:",[134,1756,1757,1760,1763],{},[37,1758,1759],{},"Strongly typed GraphQL SDL definitions.",[37,1761,1762],{},"Accurate composite type mappings.",[37,1764,1765],{},"Instant TypeScript interface generations.",[16,1767,1768,1769,1774],{},"Before an agent writes a single line of application logic, the IDP injects the exact GraphQL schema into the agent workspace, establishing ",[117,1770,1773],{"href":1771,"rel":1772},"https:\u002F\u002Fmihai.ltd\u002Fblog\u002Fend-to-end-type-safety-ai-coding-agents",[424],"end-to-end type safety for AI coding agents"," before compilation even starts. The agent operates against a verified type contract rather than speculative assumptions.",[71,1776,1778],{"className":156,"code":1777,"language":158,"meta":77,"style":77},"\u002F\u002F Auto-generated schema contract supplied to the agent by the IDP\nexport interface IssueTelemetryRecord {\n  id: string; \u002F\u002F UUIDv7\n  repoId: string;\n  turnCount: number;\n  tokensConsumed: number;\n  status: \"MERGED\" | \"REJECTED\" | \"LOOP_ABORTED\";\n  startedAt: string; \u002F\u002F ISO 8601\n  mergedAt?: string;\n}\n",[41,1779,1780,1785,1797,1813,1825,1836,1847,1870,1884,1896],{"__ignoreMap":77},[162,1781,1782],{"class":164,"line":165},[162,1783,1784],{"class":168},"\u002F\u002F Auto-generated schema contract supplied to the agent by the IDP\n",[162,1786,1787,1789,1792,1795],{"class":164,"line":172},[162,1788,176],{"class":175},[162,1790,1791],{"class":175}," interface",[162,1793,1794],{"class":190}," IssueTelemetryRecord",[162,1796,209],{"class":186},[162,1798,1799,1802,1804,1807,1810],{"class":164,"line":212},[162,1800,1801],{"class":190},"  id",[162,1803,194],{"class":175},[162,1805,1806],{"class":205}," string",[162,1808,1809],{"class":186},"; ",[162,1811,1812],{"class":168},"\u002F\u002F UUIDv7\n",[162,1814,1815,1818,1820,1822],{"class":164,"line":230},[162,1816,1817],{"class":190},"  repoId",[162,1819,194],{"class":175},[162,1821,1806],{"class":205},[162,1823,1824],{"class":186},";\n",[162,1826,1827,1830,1832,1834],{"class":164,"line":245},[162,1828,1829],{"class":190},"  turnCount",[162,1831,194],{"class":175},[162,1833,206],{"class":205},[162,1835,1824],{"class":186},[162,1837,1838,1841,1843,1845],{"class":164,"line":251},[162,1839,1840],{"class":190},"  tokensConsumed",[162,1842,194],{"class":175},[162,1844,206],{"class":205},[162,1846,1824],{"class":186},[162,1848,1849,1852,1854,1857,1860,1863,1865,1868],{"class":164,"line":293},[162,1850,1851],{"class":190},"  status",[162,1853,194],{"class":175},[162,1855,1856],{"class":631}," \"MERGED\"",[162,1858,1859],{"class":175}," |",[162,1861,1862],{"class":631}," \"REJECTED\"",[162,1864,1859],{"class":175},[162,1866,1867],{"class":631}," \"LOOP_ABORTED\"",[162,1869,1824],{"class":186},[162,1871,1872,1875,1877,1879,1881],{"class":164,"line":346},[162,1873,1874],{"class":190},"  startedAt",[162,1876,194],{"class":175},[162,1878,1806],{"class":205},[162,1880,1809],{"class":186},[162,1882,1883],{"class":168},"\u002F\u002F ISO 8601\n",[162,1885,1886,1889,1892,1894],{"class":164,"line":357},[162,1887,1888],{"class":190},"  mergedAt",[162,1890,1891],{"class":175},"?:",[162,1893,1806],{"class":205},[162,1895,1824],{"class":186},[162,1897,1898],{"class":164,"line":364},[162,1899,395],{"class":186},[92,1901,1903],{"id":1902},"_3-sandboxed-tool-registries-via-model-context-protocol","3. Sandboxed Tool Registries via Model Context Protocol",[16,1905,1906],{},"Human engineers access production infrastructure using browser sessions, SSH tunnels, and personal API tokens.",[16,1908,1909],{},"Giving an autonomous agent unrestricted shell access or broad cloud credentials creates severe security and operational risks. An agent-ready IDP restricts agent access to a curated, sandboxed tool registry implemented via the Model Context Protocol (MCP).",[16,1911,1912],{},"Each tool in the platform registry must declare:",[134,1914,1915,1918,1921,1924],{},[37,1916,1917],{},"Strict input schemas with JSON Schema validation.",[37,1919,1920],{},"Explicit read versus write designations.",[37,1922,1923],{},"Rate limits and execution timeouts.",[37,1925,1926],{},"Audit logging that captures input arguments and execution outputs.",[16,1928,1929,1930,1935],{},"If an agent needs to provision a Cloudflare KV namespace or query an internal metrics cluster, it calls a declared MCP tool rather than running unverified bash scripts. This MCP contract answers how AI coding agents should interact with an internal developer platform: every platform action goes through typed schemas instead of ad-hoc terminal commands. At the data layer, pair sandboxed MCP tools with database-enforced tenant boundaries using ",[117,1931,1934],{"href":1932,"rel":1933},"https:\u002F\u002Fmihai.ltd\u002Fblog\u002Frow-level-security-ai-coding-agents",[424],"row-level security for AI-built apps",", so a single misrouted query cannot cross tenant data.",[92,1937,1939],{"id":1938},"_4-deterministic-verification-gates","4. Deterministic Verification Gates",[16,1941,1942],{},"Human pull requests depend on asynchronous peer review. A developer opens a pull request, tags a teammate, and waits hours or days for feedback.",[16,1944,1945],{},"Autonomous agents require immediate, synchronous verification loops. If an agent must wait 24 hours for human review feedback, the cost of context switching destroys throughput.",[16,1947,1948],{},"The agentic IDP runs automated verification gates directly inside the agent workspace:",[71,1950,1953],{"className":1951,"code":1952,"language":76,"meta":77},[74],"[TASK INGESTION]\n       │\n       ▼\n[ISOLATED DB BRANCH & CONTRACT INJECTION]\n       │\n       ▼\n[AGENT AUTONOMOUS IMPLEMENTATION]\n       │\n       ▼\n[DETERMINISTIC VERIFICATION GATE]\n  ├── Unit & Contract Assertions (PASS)\n  ├── Static Type Check (PASS)\n  ├── Spec-Readiness Linter (PASS)\n  └── Churn Budget \u003C 4 Loops (PASS)\n       │\n       ▼\n[AUTOMATED PR DISPATCH & AUDIT RECORD]\n",[41,1954,1952],{"__ignoreMap":77},[16,1956,1957,1958,1960,1961,29],{},"When all deterministic gates pass, the agent opens a pull request accompanied by full telemetry traces. Human reviewers evaluate the high-level architecture instead of checking basic syntax or testing edge cases manually. Teams track cycle velocity with ",[117,1959,1625],{"href":434}," and replace story point estimation with ",[117,1962,1963],{"href":429},"spec-readiness contracts",[84,1965,1967],{"id":1966},"platform-engineering-for-the-next-decade","Platform Engineering for the Next Decade",[16,1969,1970],{},"Platform engineering is not disappearing. The target audience is changing.",[16,1972,1973,1974,1979,1980,1985],{},"Organizations that continue building visual portals for human keystroke optimization will find themselves maintaining legacy digital paper-pushing tools. Engineering organizations that evolve their IDPs into machine-readable, schema-driven platforms will enable autonomous coding agents to build, verify, and ship software with high reliability. We apply this exact pattern in production for ",[117,1975,1978],{"href":1976,"rel":1977},"https:\u002F\u002Finnovi.care",[424],"Innovi Care",", detailed in the ",[117,1981,1984],{"href":1982,"rel":1983},"https:\u002F\u002Finnovi.pro\u002Fblog\u002Fagentic-software-factory-fhir-graphpg",[424],"FHIR R5 agentic software factory case study",", where agents work directly from introspected database contracts rather than manual specs.",[16,1987,1988,1989,1994,1995,1998],{},"To audit your team's agent workflow bottlenecks and design scalable internal platforms, book a ",[117,1990,1993],{"href":1991,"rel":1992},"https:\u002F\u002Finnovi.pro\u002Fservices\u002Fsoftware-architecture-audit",[424],"Software Architecture Audit"," or explore how our advisory team at ",[117,1996,1006],{"href":1004,"rel":1997},[424]," partners with engineering leaders.",[447,2000,2001],{},"html pre.shiki code .sH3jZ, html code.shiki .sH3jZ{--shiki-default:#8B949E}html pre.shiki code .suJrU, html code.shiki .suJrU{--shiki-default:#FF7B72}html pre.shiki code .sQhOw, html code.shiki .sQhOw{--shiki-default:#FFA657}html pre.shiki code .sZEs4, html code.shiki .sZEs4{--shiki-default:#E6EDF3}html pre.shiki code .sFSAA, html code.shiki .sFSAA{--shiki-default:#79C0FF}html pre.shiki code .s9uIt, html code.shiki .s9uIt{--shiki-default:#A5D6FF}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}",{"title":77,"searchDepth":172,"depth":172,"links":2003},[2004,2010],{"id":1702,"depth":172,"text":1703,"children":2005},[2006,2007,2008,2009],{"id":1709,"depth":212,"text":1710},{"id":1733,"depth":212,"text":1734},{"id":1902,"depth":212,"text":1903},{"id":1938,"depth":212,"text":1939},{"id":1966,"depth":172,"text":1967},"2026-11-06","Human internal developer platforms prioritize visual portals and web forms. Autonomous coding agents need machine-readable contracts, ephemeral database branches, and tool registries.","IDP-001",{"type":465,"canonical":2015},"https:\u002F\u002Fretroagile.com\u002Fblog\u002Finternal-developer-platform-for-coding-agents","AGENT_IDP",{"title":1673,"description":2012},"blog\u002Finternal-developer-platform-for-coding-agents",[2020,2021,2022,476],"idp","platform engineering","coding agents","4xjq6T7fGtgRu_FIWQV1wSjBfR4JFenazINVwZCuQt0",1794728603658]