Jeff Dyer

Building and

[Jeff Dyer @compilerguy]

“Dissenters did something brilliantly simple: in an unfree country, they began to behave like free people, thereby changing the moral climate and the country’s prevailing tradition.” — Andrei Amalrik, Russian historian and dissident. via @vkaramurza

[Jeff Dyer @compilerguy]

@joesuh #2

[Jeff Dyer @compilerguy]

"The ultimate implication is that a layer between the work itself and the underlying intelligence needs to deeply understand your workflows, context, and business process." 💯

[Aaron Levie @levie]

Some good best practices here on AI token cost optimization. None of these happens though without a deep understanding of the underlying work being done in a non-abstract way.

The ultimate implication is that a layer between the work itself and the underlying intelligence needs to deeply understand your workflows, context, and business process. Now, each individual company doing this on their own is unlikely to be effective at scale, so as a consequence, this is effectively the playbook for any applied AI company right now.

By evaling the models for the applied use cases, deeply understanding the domain, having tuned UX and features for the use case, and having the ability to support adoption and change (via FDEs), allow this layer to add a ton of value. And as a result, enterprises get higher ROI because you actually can get *more* intelligence per dollar by having optimal architecture and workflows.

There will be many horizontal and vertical versions of this approach. Huge opportunity right now.

[Jeff Dyer @compilerguy]

@arvidkahl If we can get to 64 nations (roughly one quarter of all on the planet) fielding quality football teams, I am here for it

[Jeff Dyer @compilerguy]

@arvidkahl Me too!

[Jeff Dyer @compilerguy]

@arvidkahl This is why the FA Cup and March Madness are so much fun to watch. The little guys get on the big stage and sometimes step up

[Jeff Dyer @compilerguy]

@arvidkahl I'm surprised at the quality of the play of random teams against the favorites. I say let's pause at 48 teams for a cycle or two, but getting more nations of such quality is a win!

[Jeff Dyer @compilerguy]

I heard @pbakaus say raise the ceiling for craftsmen using agents. That is 100% what every AI software company should be doing. LLMs lower the floor by nature. Raising the ceiling for experts is what I am interested in

[Jeff Dyer @compilerguy]

AI doesn't change what software engineers do. It just changes how they do it. If you have ideas about what you are making, you need to stay in the loop. Same as before

[Jeff Dyer @compilerguy]

What were you doing with Saleforce that you can now do with Replit? Why didn't you hire a few good engineers a few years ago to do the same? What might go wrong now?

[Amjad Masad @amasad]

I hear a version of this almost daily.

[Jeff Dyer @compilerguy]

100%. My daily workflow is to put Claude Code in Plan Mode and talk to it for an hour or two about what I want to accomplish. With almost every turn of that planning conversation I correct Claude because it wants to do what the average code on Github wants to do. Which, by the way, is how it provides its productivity. Most of standard practices on Github are better than mine for 90% of what I want to do. But deferring to the average for a novel thing I am trying to accomplish and for which I have both expertise and an opinion is nuts. So we plan and plan and plan until I am happy with Claude's understanding of my ideas and its details of the implementation, and then let it rip. When it is done, it is usually not completely right, so we iterate. Sometimes it is completely right and that is magic

[Aaron Levie @levie]

Coding is basically the pinnacle of what you could reasonably automate with AI, and yet we still need human engineers to oversee agents for them to be effective.

The AI models are trained on an incredible amount of sophisticated code. The users are highly technical and can use the latest tools quickly. The work is “verifiable” because you can test an app. The outcomes are often removed from the quality of the code (you can have sloppy code but the app can still work). And the context for the agent is often already digitized and sitting in the codebase.

That’s an incredible amount of benefits that AI coding agents get to work with. Some of those apply to knowledge work, but most don’t in areas where the work needs to be fully reviewed to be useful, or where data isn’t as abundantly digitized. This makes the job for agents in knowledge work more complicated.

So if with all of that, engineers still remain in very high demand, the risks are going to be less than what’s perceived for other areas of knowledge work. Agents will let people do far more than they did before, but the people don’t go away.

[Jeff Dyer @compilerguy]

https://t.co/IzwvsQsc3l
Feedback welcomed.

[Jeff Dyer @compilerguy]

Every authoritarian government older than a century has failed. Only one liberal democracy has survived for as much as 250 years. Let's keep the streak alive. Be the educated and thoughtful citizen that got us here

[Jeff Dyer @compilerguy]

I don' t think I disagree with anything you write here. My point is that creativity is bounded by embodied experience. As humans, our lived experience is primarily out of reach of robots. This will be true even when we have physical robots with so called world models.

The unit distance problem, Move 37, and everyday work with Claude Code are possible because they occur in the intersection of human and machine experience.

The OP says “you don’t actually know what you want to say or what you think” when you start writing. I’d add that the thing you eventually write is often hiding partly in the unconscious mind, pulling you toward an outcome that cannot be reached through explicit reasoning alone.

[Jeff Dyer @compilerguy]

@__avik I think you hit on something, but maybe not in the way you are thinking: "...and thus may be missing sources of human creativity in some fundamental ways...". I wonder if those sources of human creativity are ultimately not able to be encoded in data as we know it. The human experiences that motivate acts of creativity are far removed from what can be encoded in a binary arithmetic. There seems to be a significant gap between the current model of artificial intelligence and what we consider to be even the most basic human intelligence. This is not to say that the current technology can't be used to create super alien intelligences. I wouldn't bet against that. I just don't see the throughline from LLMs to the human intelligence of people including very young ones.

[Jeff Dyer @compilerguy]

This model is new for us. Previously we functioned more as an agency. Too early to tell how it's going, but our new approach feels right, probably inevitable. Our OMTM is organic agent sign-ups for our MCP tools. Still looking for our first one. Definitely feels like we need to play the long game to win

[Jeff Dyer @compilerguy]

That is a big difference, determinism. High level formal language is an extension of the machine. Everything above it, whether human or agent are speculating about what the machine does. I love my coding agents. They are how I write all my code but deferring your truth to them is not work, its play at best

[Jeff Dyer @compilerguy]

@makingAISimple @arvidkahl Agreed. logic != consciousnesses && pattern matching != logic