Getting my first real exposure to multi-agent development workflows recently and… well, nothing I say will be new, but:
1) This is clearly the future.
2) Oh we’re going to buy so much inference.
3) Barely feels like software engineering but I end the day intellectually spent.
1) This is clearly the future.
2) Oh we’re going to buy so much inference.
3) Barely feels like software engineering but I end the day intellectually spent.
Oct 15, 2026 · 4:03 AM UTC
172
71
2,916
121,730
4) I think every estimate of aggregate speed up or increase in total software shipped will be an underestimate. This is wild. It is wild even compared to the single-threaded LLM-assisted modern coding tool baseline.
5
6
423
15,857
5) Remember when dinosaurs walked the early and software engineers actually read all the code they were responsible for?
6) It is deeply weird to, in the middle of extremely productive work, have a long argument with a coordination process attempting to argue with a subagent.
6) It is deeply weird to, in the middle of extremely productive work, have a long argument with a coordination process attempting to argue with a subagent.
8
3
356
16,899
@patio11 I've been running Claude Code in Ultracode mode and it is wild
6
0
36
7,339
@patio11 It’s like a cross between being an engineering manager and an architect
2
0
9
2,288
@patio11 Far more akin to managing an eng org/system, only no emotional issues from the team to deal with and the set of what you can trust LLMs to handle without strict rules is weirder than with humans...
1
0
3
255
@patio11 Shipping the software is easier than ever but getting people to care about it for most is a different story.
0
0
2
746
@patio11 Once agents run in parallel, each one needs its own isolated environment, so sandbox compute ends up sitting right next to inference in the bill.
0
0
0
170
@patio11 putting yourself in the frame of working with agents is directionally wrong. we are a bottleneck and will rightfully be abstracted away.
directing AI will look like directing your 401k today...say by 2030
directing AI will look like directing your 401k today...say by 2030
1
0
21
3,574
@patio11 Alright so can you ship some of that software please, I'm still waiting for even one thing I need to be finished.
1
0
13
2,760
@patio11 I feel we are basically all technical product managers now. Being able to envision what needs building and have the deep technical knowledge to know if its being built well and steer it accordingly, super powerful mix.
0
0
3
538
@patio11 the inference bill on this stuff is going to be wild
0
0
2
1,067
@patio11 コードを書く時間が減っても、複数の出力を見比べて判断する負荷は残りますよね。エージェントごとの役割と確認ポイントが固まるまでは、むしろ頭を使いそうです。
0
0
0
348
@patio11 We automated the typing and immediately invented meetings between processes.
0
0
0
11
@patio11 so, the b-roll chaos is real. had that exact nonsense pop up yesterday. baking tutorial, football cuts, like the ai just grabs whatever’s in the 'sports' folder and calls it a day. kinda das ist working but also not . anyway
0
0
0
342
@patio11 “but I end the day intellectually spent” highly relatable
0
0
2
425
@patio11 I think there's an upper limit on the consumption side.
Even if everyone is paid by the government to consume for 8 hours a day there's not enough human biomass for what all the data centers worth of inference can produce.
Even if everyone is paid by the government to consume for 8 hours a day there's not enough human biomass for what all the data centers worth of inference can produce.
0
0
0
388
@patio11 it’s so extremely not software engineering and it’s still tiring and you get so so much done.
0
0
0
176
