Sabitlenmiş Tweet
Phil Wait
9.1K posts
Phil Wait
@philomglol
two thirds hydrogen
Melbourne, VIC Katılım Nisan 2007
1.1K Takip Edilen357 Takipçiler
Phil Wait retweetledi
basically: anthropic sneakily turned down how hard claude thinks before editing code, changed the default from "high" to "medium" effort, and hid the reasoning from session logs. all without telling users.
an amd director had 7k sessions of telemetry to prove the degradation was real and measurable (not just vibes). anthropic admitted to the changes. there's a workaround (use "/effort max"). the uncomfortable part is most users had no data to notice it happened at all.
ℏεsam@Hesamation
AMD Senior AI Director confirms Claude has been nerfed. She analyzed Claude's session logs from Janurary to March: > median thinking dropped from ~2,200 to ~600 chars > API requests went up 80x from Feb to Mar. less thinking and failed attempts meaning more retries, burning more tokens, and spending more on tokens > reads-per-edit dropped from 6.6x → 2.0x. model stops researching code before touching it. > model tried to bail out or ask "should i continue" 173 times in 17 days (0 times before March 8). > self-contradiction in reasoning ("oh wait, actually...") tripled. > conventions like CLAUDE.md get ignored because there's less thinking budget to cross-check edits > 5pm and 7pm PST are the worst hours, late night is significantly better. this means the thinking allocation is most likely GPU-load-sensitive.
English
Phil Wait retweetledi
the CTO after the VP of Sales says “I’ve been Token-maxxing” and asks for access to the production codebase
English
Phil Wait retweetledi
Phil Wait retweetledi
Phil Wait retweetledi
Phil Wait retweetledi
Proud with @UNSWRNA to have been involved & making the mRNA-LNP for Rosie. There are nuances here that the thread below misses but nevertheless, the intersection of RNA technology, genomic & AI poses an opportunity to change the way do medicine and make access more equitable 1/8
Greg Brockman@gdb
How AI empowered Paul Conyngham to create a custom mRNA vaccine to cure his dog’s cancer when she had only months to live. The first personalized cancer vaccine designed for a dog:
English
Phil Wait retweetledi
When a team fully activates on agents and each engineer is shipping 10+ PRs a day minimum, the entire traditional SDLC collapses.
Code review backlog, keeping up with docs, keeping CI and the merge queue fast, planning, visibility, etc.
You end up redesigning the whole SDLC.
English
Phil Wait retweetledi
Phil Wait retweetledi
On one end, the Anthropic team is a massive user of AI to write code (80%+ of all code deployed is written by Claude Code). They ship amazingly fast.
On the other hand, seeing these beyond terrible reliability numbers suggests there might be a downside to all this speed:
English
Phil Wait retweetledi
there's an AWS outage in me-central-1 because it got bombed
English
Phil Wait retweetledi
Code migrations are one of the most expensive, soul-crushing line items in enterprise engineering. A typical framework migration (React class to hooks, Jest to Vitest, Angular to React) takes a team of 3-5 engineers somewhere between 2-6 months. At $150-200/hr loaded cost, that’s $200K-$500K per migration for a mid-size codebase. And most companies have a backlog of 5-10 migrations they’ve been avoiding for years because the math never works.
/batch rewrites the math entirely.
Each agent gets its own git worktree. Full isolation. It writes the code, runs the tests, and opens a PR. Dozens of these running in parallel means what used to take a team a quarter now takes an afternoon of review.
The serialization bottleneck is gone. Migrations have always been constrained by the fact that one human can only touch one file at a time, needs context on the codebase, and gets fatigued by the repetition. The work itself was never intellectually hard. It was volumetrically hard. And volume is exactly what parallelized agents solve.
This changes how engineering leaders think about technical debt. Every CTO has a spreadsheet of migrations they’ve been deferring because the ROI never justified pulling engineers off feature work. /batch turns those from “someday” projects into Tuesday afternoon tasks.
Anthropic is building the infra layer that makes AI agents useful for real engineering work, not demos. And /batch is the clearest signal yet that they understand the actual bottleneck: developers don’t need help writing new code nearly as much as they need help moving old code forward.
Boris Cherny@bcherny
In the next version of Claude Code.. We're introducing two new Skills: /simplify and /batch. I have been using both daily, and am excited to share them with everyone. Combined, these kills automate much of the work it used to take to (1) shepherd a pull request to production and (2) perform straightforward, parallelizable code migrations.
English
@PAXAus Surely @TheSpiffingBrit would make a terrific commonwealth themed display
English
If you could 🪄SUMMON✨ your ultimate dream guest to #PAXAus... who are you calling in? 👀
English
Phil Wait retweetledi
having cats is wild
there's a whole ass obligate carnivore apex predator with a will of his own sitting on my couch staring at me like this but i love him very much and his name is salad
English
Phil Wait retweetledi
now that our PM knows how to vibecode, feels great to just say: "good idea, ping me for a code review" every time he brings me up an idea I don't feel like implementing
English
Phil Wait retweetledi
Activist investor Palliser Capital sent a letter to $7B Japanese toilet maker Toto and said it was “the most undervalued and overlooked AI memory beneficiary”.
Toto known for its bidet toilets but the expertise in ceramics is crucial for memory manufacturing.
Per FT, “Toto’s chuck technology uses ceramics designed to remain stable at very low temperatures, helping hold silicon wafers firmly during chip production. That makes it relevant to cryogenic etching, which is expected to grow as memory chips become more layered and complex.”
Palliser believes Toto has a 5-year moat on the technology and should expand the operation.
Advances ceramics already make up 40% of Toto’s operating profit while being only <10% of revenue.
Toto is up +60% over the past year on their development.
English
Phil Wait retweetledi
Phil Wait retweetledi
Phil Wait retweetledi
One nine.
One.
Nine.
Swizec Teller@Swizec
Leadership: "AI is so good it writes almost all our code" Status dashboard:
English