Vol. 01 · Field notes on AI & dev tools49 fixes · 16 benchmarks · 11 AI · 6 deploy · 5 agents · 8 database
Hi, I'm Nilay.
I test AI & dev tools, then write down what actually happened.
I'm a developer. I try new models, frameworks and tools on real projects, and keep notes: what it cost, how fast it was, what broke and how I fixed it. This site is those notes, cleaned up.
start here: entry No.095
On my desk now
Next up: auth & payments.
Rules of this notebook
- 1.Every error reproduced first.
- 2.Exact versions, written down.
- 3.Real numbers, not adjectives.
- 4.Failures stay in the post.
- 5.Nobody pays for a verdict.
— N.K.
Fresh off the bench
- Supabase CLI 3 now points db push at your local stack →
- TypeScript 7.1 beta fixed none of the 9 breaks from 7.0 →
- bun check: 3.7x faster than tsc 7 on Zod, same errors →
- next: auth & payments errors (Stripe, Clerk & Auth.js), same rules
On the bench lately
AI models, Coding agents, LLM APIs, Next.js, Astro, Cloudflare, Vercel, Node.js, Android, Automation, Dev tools, Open-source.
Part 01 · 16 entries
Tested
Dev tools, measured for real: speed, size, what broke.
Open →
Part 02 · 49 entries
Fixes
Exact error message, why it happens, the fix that worked, versions tested.
Open →
Part 03 · 11 entries
AI
AI SDKs and APIs, run for real: the errors they throw, reproduced and fixed.
Open →
Part 04 · 6 entries
Deploy
Deploy and hosting errors, reproduced on a real deploy: what the platform really did.
Open →
Part 05 · 5 entries
Agents
Coding agents run for real: Claude Code, Codex, Gemini CLI and MCP, the errors they hit and the fix.
Open →
Part 06 · 8 entries
Database
Supabase, Postgres and Prisma errors, reproduced on a real local stack: the error code, the cause, the fix.
Open →
The log
Follow by RSS- No.095
Supabase CLI 3.0.0 tested: what breaks coming from v2
On release day, Supabase CLI 3.0.0 started the same local stack as 2.122.1, and row-level security behaved identically. What changed is the targeting: db push and gen types with no flags now run against the local stack and print Target: local stack, where v2 failed in the same unlinked project. --linked still works but warns, db remote commit is gone, an old [inbucket] section in config.toml now stops every command, and an invalid SUPABASE_* variable is rejected by name.
TESTED3 min - No.094
Python CERTIFICATE_VERIFY_FAILED, tested on 3.12 to 3.15
certificate verify failed: unable to get local issuer certificate, or Basic Constraints of CA cert not marked critical on Python 3.13+. Each cause tested.
> requests.exceptions.SSLError: HTTPSConnectionPool(host='pypi.org', port=443): Max retries exceeded with url: /simple/
FIXED3 minWindows - No.093
Fix: bigint Failed to load bindings, pure JS will be used
bigint: Failed to load bindings, pure JS will be used. Why npm 12 shows it on every install, whether it matters, and how to get the native build back.
> bigint: Failed to load bindings, pure JS will be used (try npm run rebuild?)
FIXED3 minWindows - No.092
TypeScript 7.1 beta tested: ts-jest, ESLint, NestJS, Angular
On TypeScript 7.1.0-beta, ts-jest, typescript-eslint, ts-node and TypeDoc failed exactly as on 7.0, and so did NestJS, Angular, vue-tsc, svelte-check and astro check. The beta does ship a compiler API, but at new entry points (typescript/sync and typescript/async) that no tool uses yet. The side-by-side setup with TypeScript 6 fixed every one I re-ran.
TESTED4 min - No.091
Fix: npm error code EALLOWREMOTE (and EALLOWGIT), tested
npm error code EALLOWREMOTE: Fetching packages of type remote have been disabled. Every case reproduced on npm 12, from npx to npm ci, with each fix.
> npm error code EALLOWREMOTE
FIXED4 min - No.090
Node 26 error: libatomic.so.1 cannot open shared object file
node or npm fails with error while loading shared libraries: libatomic.so.1 after moving to Node 26. Tested on 9 Linux images, Docker and nvm, with fixes.
> node: error while loading shared libraries: libatomic.so.1: cannot open shared object file: No such file or directory
FIXED4 min - No.089
module 'openai' has no attribute 'ChatCompletion', tested
AttributeError: module 'openai' has no attribute 'ChatCompletion', or 'no longer supported in openai>=1.0.0'. Five SDK versions tested.
> AttributeError: module 'openai' has no attribute 'ChatCompletion'
AI3 min - No.088
Responses.create() unexpected keyword argument: new names
max_tokens, messages, response_format, seed: their names in OpenAI's Responses API, tested. Plus the 'proxies' TypeError from httpx 0.28.
> TypeError: Responses.create() got an unexpected keyword argument 'max_tokens'
AI6 min
Post card · Every Sunday
What broke this week.
One email on Sundays: the Node, TypeScript, Python and AI SDK releases that broke builds that week, each one reproduced, with a link to the fix that worked. Never shared, and you can leave any time.
— Nilay