Drew Breunig
16K posts
Drew Breunig
@dbreunig
Writing about and working on AI, DSPy, geo, and data.
Bay Area Katılım Mart 2008
1.2K Takip Edilen9.3K Takipçiler
Every time we get a new technology, we shoehorn its capabilities into a form factor that fits our infrastructure, culture, and habits. We use it in a way we currently know. This is never the final form.
Here's one of the first automobiles:
Simon Willison@simonw
The idea of "AI employees" feels so short-sighted to me - both disrespectful to humans and a complete misunderstanding of what these tools can do and how to best put them to work You may as well start adding Excel spreadsheets to your org chart
English
Drew Breunig retweetledi
The idea of "AI employees" feels so short-sighted to me - both disrespectful to humans and a complete misunderstanding of what these tools can do and how to best put them to work
You may as well start adding Excel spreadsheets to your org chart
English
@MaximeRivest Interesting. I’d count the investment in the creation of those skills/etc as part of the effort. But I still think the risk is there if your inputs diverge.
English
@dbreunig can specialized workflows / prompts skip the valley? or maybe the better question: how often / easy can a specialized workflows, prompts, model skip the valley for that 'type' of task?
English
This feels right. Slop occurs because people think AI's first shot is sufficient. It takes many turns of honing to exceed the communicative quality of your initial input.
If you aren't going to spend the time and turns, send your original prompt as the email.
English
@davidcrawshaw Duplication is probably ideal when you have many agents working at once.
English
Everything we have done statically has been half measures because verification is too expensive. It’s cheap now so ditch it all.
Duplication is fine when you have agents constantly reading and synchronizing the code base.
English
Some guesses about the future of programming languages, given agents write and humans read:
- untyped
- out of band formal verification
- test harness runs code in a simulator vm to instrument the os
- less expressive, more copying
English
We are now at the point in our household where I am printing out papers so our 10 year-old can attempt to break small LMs.
Drew Breunig@dbreunig
This is fun.
English
Taken together with the weak stand-alone Sonnet 5 results, is the current low/med/high setup becoming obsolete for coding tasks? Is the right pattern a giant planner and a capable subagent, with the middle size dropping out?
Artificial Analysis@ArtificialAnlys
GPT-5.6 Sol and Luna are ahead of Terra at every point on the Intelligence vs Cost per Task chart. GPT-5.6 Luna stands out as a particularly cost efficient model Charting the Artificial Analysis Intelligence Index shows the trade-off between intelligence and Cost per Intelligence Index Task. Across reasoning efforts, each GPT-5.6 model pushes past GPT-5.5 on the Pareto frontier (excluding non-reasoning). However, Luna and Sol are always ahead of Terra. This means for any Terra effort level, there is a Luna or Sol effort level that is more intelligent at no extra cost, or as intelligent at lower cost.
English
Linear's app is terrible but its data structure is incredible. Perfect for agents.
English
@MaximeRivest C’mon, you could have put that in disappearing handwritten script
English
Wow, gpt-5.6-sol made me a android app that turns my phone into microphone for my eink tablet!
There is something special from getting words on a page using my voice and editing it using a pen and paper-feel.
sound on!
ps: no harry potter reference in this one, sorry!
the reason that I am exploring all these elements in eink is that I am convinced that eink+ai can lead to a much much better type of human computer interaction.
English
English
@dbreunig @stewartbrand @cloudnativegeo I’ve been talking to Jed about coming for a while now! Maybe should bite the bullet, as it were. Which one are you joining?
English
Great blog from fellow FOO Camper @dbreunig on @stewartbrand's Pace Layers framework applied to the AI ecosystem.
Curious where you'd put earth / planetary data here, Drew, as the nature layer also speeding up significantly now.
dbreunig.com/2026/07/03/ai-…
English
English
@denisehearn_ @stewartbrand Earth observation (and more) data is getting way more dynamic… But it’s still probably in the decades or years due to the instrumentation and market requirements. We’re mining what we have already (Landsat, SAR), data accrues steadily but in real time.
English
Drew Breunig retweetledi
Seeing someone’s "own" write up being full of Claudisms
Just give me the bullet points used to generate the article, they are far more interesting than the overly complicated slop
English