Did the agents actually ship it?

29 stories

• Vendor-published Language migration brownfield

  GitHub ports the Copilot agent runtime from TypeScript to 800K+ lines of Rust, mostly by one engineer

  GitHub (Microsoft) · 2026-09-16

  • 832,378 lines Production Rust written
  • ~430,000 lines TypeScript passed through the port (author estimate)
  • ~14.5 weeks Porting window

  GitHub Copilot app, GitHub Copilot CLI

• Vendor-published Test migration brownfield

  Asana finishes a five-year Enzyme → React Testing Library migration in two weeks with Codex

  Asana · 2026-08-19

  • ~2 weeks Calendar time
  • ~1.5 weeks Engineering effort
  • Up to 4 Parallel agents

  OpenAI Codex

• Independent Research study

  Harvard study of 718 firms: AI agents raise code volume but not shipped work, review becomes the bottleneck

  718 firms (Jellyfish data) · 2026-08-04

  • 718 Firms
  • ~300 million Work events
  • +30% Lines of code after agent adoption

  AI coding agents (various), AI coding assistants (various)

• Independent Cost governance

  Uber caps AI coding spend at $1,500 per employee per tool after burning its annual AI budget in 4 months

  Uber · 2026-06-02

  • $1,500 per employee, per tool Monthly cap
  • 4 months Time to exhaust annual AI budget

  Claude Code, Cursor

• Vendor-published Benchmark brownfield

  Legacy-Bench: frontier agents pass only 16.9–42.5% of COBOL, Java 7 and Fortran tasks

  Factory · 2026-04-01

  • 16.9%–42.5% Pass rate range (12 model–agent combos)
  • >70% Same models on Terminal-Bench 2 / SWE-bench Verified
  • 42.5% Best overall (GPT-5.3-Codex)

  Droid, Codex, Gemini CLI

• Independent Research study

  METR's late-2025 follow-up: raw data hints at speed-up, but developers refusing to work without AI breaks the study

  METR (57 open-source developers) · 2026-02-24

  • 57 Developers
  • 143 Repositories
  • 800+ Tasks

  AI coding tools (various, incl. agentic tools)

• Vendor-published New product (research) greenfield

  16 parallel Claude agents build a 100,000-line Rust C compiler that builds Linux, for ~$20,000

  Anthropic (Nicholas Carlini) · 2026-02-05

  • Just under $20,000 API cost
  • ~2 weeks, nearly 2,000 Claude Code sessions Duration
  • 100,000-line compiler Output

  Claude Code, Claude Opus 4.6

• Vendor-published New product (research) greenfield

  Cursor's agent swarm writes a 1M+ line browser in about a week, and critics find it barely compiles

  Cursor (Wilson Lin) · 2026-01-14

  • Close to a week Duration
  • 1M+ lines across ~1,000 files Output
  • Hundreds of concurrent workers Agents

  Cursor (multi-agent research harness), GPT-5.2, GPT-5.1-Codex

• Vendor-published Build optimization / defect remediation brownfield

  Cisco uses Codex across 15+ repos to cut build times ~20% and speed up C/C++ defect fixing

  Cisco · 2026-01

  • 1,500+ per month Engineering hours saved
  • ~20% reduction Build time
  • 10–15× increase Defect resolution throughput

  OpenAI Codex, Codex CLI

• Practitioner (own blog/paper) Language port / new library greenfield

  Simon Willison ports an HTML5 parser from Python to JavaScript with Codex CLI in about 4.5 hours

  Simon Willison (independent developer) · 2025-12-15

  • ~4–4.5 hours Idea to finished library
  • 8 Prompts
  • 9,000 lines of JavaScript, 43 commits Output

  OpenAI Codex CLI, GPT-5.2

• Independent New library greenfield

  Emil Stenström builds JustHTML, a pure-Python HTML5 parser passing all 9,200+ html5lib tests, with Copilot agent mode

  Emil Stenström (independent developer) · 2025-12-14

  • A couple of months of off-hours Time
  • ~3,000 lines (plus ~11,000 lines of tests) Implementation code
  • All 9,200+ html5lib-tests passing

  GitHub Copilot (agent mode), Claude Sonnet 3.7, Gemini 3 Pro, Claude Opus

• Practitioner (own blog/paper) Migration brownfield

  Spotify's background coding agent merges 1,500+ migration PRs with 60–90% time savings

  Spotify · 2025-11

  • 1,500+ Agent PRs merged to production
  • 60–90% vs writing by hand Time saved on these migrations
  • Hundreds of developers Users

  Claude Code, Claude Agent SDK, Internal CLI (Honk)

• Practitioner (own blog/paper) Feature brownfield

  Mitchell Hashimoto ships a non-trivial Ghostty feature with Amp in 16 sessions for $15.98

  Ghostty (Mitchell Hashimoto) · 2025-10-11

  • 16 Agent sessions
  • $15.98 Token spend
  • ~8 hours, over 3 days from first to last commit Hands-on time (author's estimate)

  Amp

• Practitioner (own blog/paper) Modernization brownfield

  Claude Code ports a 25-year-old Linux tape driver from kernel 2.4 to 6.8 over two evenings

  Dmitry Brant (independent developer) · 2025-09-07

  • Two evenings Time
  • 2.4 → 6.8 Kernel version jump
  • Probably several weeks (would have meant learning 25-year-old kernel development) Author's by-hand estimate

  Claude Code

• Independent Research study brownfield

  METR RCT: experienced open-source developers took 19% longer with early-2025 AI tools

  METR (16 open-source developers) · 2025-07-10

  • 16 Developers
  • 246 Tasks
  • 19% longer with AI Measured effect

  Cursor, Claude 3.5/3.7 Sonnet

• Practitioner (own blog/paper) New product greenfield

  Solo developer ships a 20,000-line native macOS app, writing fewer than 1,000 lines by hand

  Indragie Karunaratne (independent developer) · 2025-07-01

  • ~20,000 lines Codebase size
  • <1,000 lines (author's estimate) Written by hand
  • 2,000 lines of Python, written by Claude in a few hours of iteration Release automation script

  Claude Code, Claude Sonnet 4, Claude Opus 4

• Practitioner (own blog/paper) New library greenfield

  Cloudflare's Kenton Varda prompts an OAuth 2.1 provider library out of Claude Code in about five days

  Cloudflare · 2025-06-03

  • ~5 days of heavy work (with other duties), plus minor commits over the month Time with AI
  • ~5,000 lines incl. comments, tests and docs Library size
  • 2–5x longer ("a few weeks, maybe months") Author's by-hand estimate

  Claude Code

• Practitioner (own blog/paper) New product greenfield

  Pieter Levels builds a browser flight sim with Cursor in ~3 hours; it hits $1M ARR in 17 days

  Pieter Levels (indie maker) · 2025-03-11

  • ~3 hours (author's estimate) Time to first playable version
  • $87,000 MRR (~$1M ARR) Revenue after 17 days
  • 320,000 Players after 17 days

  Cursor

• Practitioner (own blog/paper) Test migration brownfield

  Airbnb migrates ~3.5K Enzyme test files to React Testing Library in 6 weeks with LLMs

  Airbnb · 2025-03

  • ~3,500 Test files migrated
  • 6 weeks Actual duration
  • 1.5 years of engineering time Manual estimate

  Frontier LLMs (unnamed)

• Practitioner (own blog/paper) Large-scale refactor brownfield

  Google Ads uses LLMs to migrate 32-bit IDs to 64-bit in a 500M+ line codebase

  Google · 2025-01

  • 500M+ lines Codebase size
  • 80% Code changes fully AI-authored (landed CLs)
  • ~50% reduction (engineer-reported estimate) Total migration time

  Gemini (fine-tuned, internal toolkit)

• Practitioner (own blog/paper) Test migration brownfield

  Google migrates 5,359 test files from JUnit3 to JUnit4 in 3 months with LLMs

  Google · 2025-01

  • 5,359 Files migrated
  • 149,000+ Lines modified
  • 3 months Duration

  Gemini (fine-tuned, internal toolkit)

• Practitioner (own blog/paper) Library migration brownfield

  Google estimates ~89% time saved on Joda-Time → java.time migration clusters

  Google · 2025-01

  • ~89% Time saved on small clusters (estimate)

  Gemini (fine-tuned, internal toolkit)

• Vendor-published Migration brownfield

  Nubank uses fine-tuned Devin to split a 6M-line ETL monolith, claiming 8–12x engineering-hour efficiency

  Nubank · 2024-12

  • 6M+ lines, ~100,000 data-class implementations Codebase
  • 18 months, 1,000+ engineers Original estimate
  • 8–12x Engineering-hour efficiency

  Devin (fine-tuned)

• Vendor-published Version upgrade brownfield

  Amazon upgrades tens of thousands of production apps to Java 17 with Amazon Q Developer

  Amazon · 2024-08-01

  • Tens of thousands (Java 8/11 → 17) Applications upgraded
  • 4,500+ years Development work saved (estimate)
  • $260M from performance improvements Annual cost savings (estimate)

  Amazon Q Developer (code transformation agent)

• Vendor-published Rollout / adoption

  Stripe rolls Claude Code out to 1,370 engineers with a zero-configuration binary

  Stripe

  • 1,370 Engineers with Claude Code enabled
  • 2–3 months Time to produce signed enterprise binary
  • 6 Coding assistants enabled at Stripe

  Claude Code

• Vendor-published Language migration brownfield

  Stripe migrates 10,000 lines of Scala to Java in four days

  Stripe

  • 10,000 Lines migrated
  • 4 days Actual duration
  • 10 engineering weeks Manual estimate

  Claude

• Vendor-published Modernization brownfield

  Altisource modernizes 350,000 lines of legacy Java with Amazon Q Developer

  Altisource

  • 350,000+ lines Legacy Java modernized
  • +25% (one modernization team, 'up to') Developer productivity
  • ~160 → 200 Story points (3-month avg per engineer)

  Amazon Q Developer

• Vendor-published Rollout / adoption

  DTCC reports 40% higher developer throughput in an Amazon Q Developer proof of concept

  DTCC

  • +40% average Developer throughput (POC)
  • −30% average Code defects (POC)
  • +5% Security scores across repos

  Amazon Q Developer

• Vendor-published Rollout / adoption

  Palo Alto Networks onboards 2,000 developers to a Sourcegraph Cody + Claude coding assistant in 3 months

  Palo Alto Networks

  • 2,000 in 3 months Developers onboarded
  • 150 developers Pilot group
  • Up to +40%, average 20–25% Coding productivity

  Sourcegraph Cody, Claude (Amazon Bedrock)

No stories match those filters yet. Know one? Submit it.