GitHub has published a first-party engineering account on github.blog (retrieved directly in this search, no longer echo-only) of rewriting the shared runtime behind Copilot CLI, the Copilot app, and the Copilot SDK from TypeScript/Node.js to Rust using Copilot itself: agents wrote most of the code, producing more than 800,000 lines of production Rust across 128 pull requests over roughly 14.5 weeks, shipped incrementally with each PR swapping one TypeScript component for a Rust implementation behind a thin shim and deleting the old code atomically. Secondary coverage (The Register) reports ~$120K in AI token spend plus a few weeks of developer time and 'a few dozen' regressions — mildly conflicting with GitHub's own 'few inevitable regressions' — and says work was routed across multiple models (GPT-5.6 Sol, Claude Opus 4.8) by task strength. The first-party account's most quotable details: agents did ~10x more exploration (reading, searching, diagnostics) than code mutation, the human engineer's role was largely supervisory (review/testing/CI), and the motivation was escaping Node/V8 process overhead rather than Rust enthusiasm — the new runtime exposes a C ABI for in-process FFI embedding, with performance reportedly improved by orders of magnitude. It remains a self-reported account with no independent verification of behavior preservation or true completion, but it now pairs with Google's Bug Hunters report on AI-assisted C/C++→Rust dependency rewrites as a second instance of the same pattern: coding assistants doing substantial systems-language migrations.
GitHub's directly retrieved first-party account is a dated-receipts convergence with AI Legacy Takeover: one component per PR swapped behind a thin shim with atomic deletion is Strangler Fig cutover verbatim, executed at 800K-line scale with a supervisory human, task-strength model routing (Scout–Senior shape), and wholesale component replacement rather than in-place surgery — external validation of the framework's premise from the most authoritative possible source. The publishable gap is equally load-bearing: the account documents no characterisation harness or convergence metrics, and 'few inevitable regressions' versus The Register's 'a few dozen' is exactly where his Verify phase predicts unmeasured trouble — evidence the industry is rediscovering Build/Promote while skipping governance, which sharpens rather than threatens his differentiator.
ip:framework.ai-legacy-takeoverip:concept.strangler-fig-patternip:concept.characterisation-testingip:concept.surgery-problemip:framework.scout-senior-splitip:concept.token-economicsradar:checkly-agentic-go-rewriteradar:anthropic-claude-code-migrationsradar:canonical-c-to-rust-projectradar:bitcoin-rs-ai-verified-node
queries asked of Scott's wikis
- AI legacy takeover replacing existing software
- surgery problem behavior-preserving legacy rewrite
- governed agent migration verification and review gates
- multi-agent orchestration child sessions worktrees inter-agent communication
- model routing heterogeneous LLM task assignment
- agentic coding token economics cost per line
now 0 pts/hpeak 2 pts/hcomments 0/hpeers p14momentum: steady2 platformsage 591h
points/hour across evidence · reading as of 2026-10-12 02:59:37.977291+11:00 · deterministic, not a model opinion
2026-09-29T03:12:23Z
grounded: converges/high — GitHub's directly retrieved first-party account is a dated-receipts convergence with AI Legacy Takeover: one component per PR swapped behind a thin shim with at
2026-09-29T03:02:59Z
Google's Bug Hunters report on AI-assisted C/C++→Rust dependency rewrites gives the case its first genuinely independent line of evidence, upgrading it from a single-source echo of GitHub's account to a corroborated pattern: two first-party organizations now report coding assistants doing substantial systems-language migrations. It corroborates the practice, not GitHub's specifics — scope, the $120K figure, and verification methods remain unverified — and attention stays cold (~1.8 pts/h residual on aged submissions; the 87th-percentile peer reading reflects a quiet cohort, not real speed), so low heat.
2026-09-29T02:30:37Z
evidence attached: hn.story.49884237 — Google's Bug Hunters reporting AI-assisted C/C++→Rust dependency rewrites is independent corroboration that coding assistants are being used for large-scale systems-language migrations, not just GitHub's own runtime.
2026-09-23T23:25:36Z
The sixth HN submission is another low-traction duplicate of the same story with no discussion, and engagement across all evidence has decayed to zero — the episode is settled attention-wise without ever producing independent corroboration, engineering detail, or confirmation of the $120K cost claim or migration scope. The case stays a single-community echo of a first-party account, not evidence of governed agent-driven migrations.
2026-09-22T11:23:45Z
evidence attached: hn.story.49799004 — shared external link with case evidence
2026-09-20T13:40:37Z
The fifth HN submission repeats the same migration account without adding engineering evidence or expanding into a new community. It does not strengthen the case for reliable systems-language migrations or clarify the previously reported cost.
2026-09-20T13:22:11Z
evidence attached: hn.story.49775125 — shared external link with case evidence
2026-09-20T11:21:31Z
The latest HN submission introduces a $120K cost claim and a quoted example of migration complexity, but neither comes with a retrieved engineering account or independent validation. This adds questions about migration economics, not evidence of successful, cost-effective or behavior-preserving replacement.
2026-09-20T11:21:18Z
evidence attached: hn.story.49773998 — Independent HN coverage materially contextualizes Microsoft's reported agent-assisted Copilot runtime migration to Rust and its claimed cost.
2026-09-17T19:22:57Z
The third submission repeats the same article without adding engineering evidence or independent corroboration. The case remains an unverified migration example, not evidence that Copilot reliably executes substantial systems rewrites.
2026-09-17T19:21:49Z
evidence attached: hn.story.49745245 — shared external link with case evidence
2026-09-17T11:26:28Z
The new attachment is another submission of the same article, not independent corroboration or a migration result. Discussion adds opinions about Copilot and implementation choices but no evidence that warrants stronger claims or continued elevated attention.
2026-09-17T11:21:45Z
evidence attached: hn.story.49738939 — shared external link with case evidence
2026-09-17T01:30:30Z
grounded: converges/low — GitHub’s reported Copilot-assisted runtime migration narrowly converges with AI Legacy Takeover’s premise that AI can help replace existing software, but curren
2026-09-17T01:27:59Z
case created — A distinct first-party systems-migration account is relevant, but title-only evidence does not support the scout's stronger acceleration and validation claims.