I talk about #programming, #ReactJS, #TypeScript, #Tesla, and gamer culture.
Manhattan, NY
Joined June 2009
- Tweets1.1K
- Following144
- Followers208
- Likes932
The fact that Google cites DeepSWE as a benchmark after we
1) Know they have many flawed tests
2) Are made public
Should indicate what you should expect from the model, which is more of useless garbage from Google.
For hard engineering problems:
- Run with Opus 5.5
- Advise (when needed) with Fable 5.1 subagent
- Explore with Grok 4.7 fast
- Review with Sol 6
For straightforward "PM" type work: Just run with Grok.
OpenAI have done interesting stuff; but the latest releases are disappointing by them and ANT has gone from disappointing releases to exciting ones with Opus 5.5. I predict this trend will now reverse.
The age of models is over. Welcome to the age of harnesses.
Introducing Projects, a new way of working in Cursor.
Rather than creating a chat for every task, you work with a coordinator agent in a single, persistent thread.
Like @bot, your agent is always on, proactively manages work with subagents, and improves over time.
Nathaniel Tucker retweeted
Introducing Projects, a new way of working in Cursor.
Rather than creating a chat for every task, you work with a coordinator agent in a single, persistent thread.
Like @bot, your agent is always on, proactively manages work with subagents, and improves over time.
Nathaniel Tucker retweeted
React 19.3 is now available!
This release makes View Transitions and Fragment Refs stable, and adds browser(), Trusted Types support, and Context in Server Components.
react.dev/blog/2026/09/09/re…
Currently the strongest SWE workflow is Cursor with Fable 5.1 plan mode, Sol advisarial review, then execute with Sol. (If cost is not object - this gives the most consistently best results in most cases.)
My prediction is Grok 4.7 will likely dethrone Sol for code writing.
Astra probably most disappointing model release so far. Worse models have come out, but noone yet expects Google to produce something useful.
Next step for @bot please:
- Let us vibe code the UI
- Make stronger data primitives like threading
- Allow bots to output and update UIs in-line (like cursor can do). I really want more easily parseable reports in there.
I've been liberally increasing my Grok usage to try to fill that bar and it just keeps shrinking after :P. Because Grok is so good now I finally got to 20% usage for this period but now it's down to 10%.
The thing missing from @bot is bespoke generated UI. Viewing everything as text is way too much work for my feeble human brain to parse.