2026-05-27
Three threads from today: the economics of AI gains, the structure of leadership clarity, and why the model is no longer the product.
Dax Raad / OpenCode
A lot of the job has become objectively easier with AI — but why does it feel like I'm still having to think as hard as I ever did?
Tooling Economics
The Hard Parts Remain
AI compresses the implementation cost — diffs get written, refactors run across a whole codebase. But taste, architecture, product understanding, and knowing what to build: those costs don't move. The job is objectively easier in many places and still demands as much thinking.
Same Work, Faster
The natural way for engineers to cash out AI tooling gains is time savings — the same tickets, done faster. Converting those gains to higher-leverage output requires deliberately redesigning incentives, not just upgrading prompts.
Cleanup Gets Cheap
Asking an agent to implement a new pattern everywhere across a codebase is cheap now. Tech debt cleanup shifts from a rare capital project requiring a sprint to ongoing maintenance anyone can trigger.
Agents Need Guardrails
Domain-driven design and verbose patterns went out of fashion because they were tedious to type. They're coming back — because agents behave like fast junior engineers who need the same structural constraints those patterns provide.
Communication
Most updates fail not because the work is bad but because they lead with activity instead of outcome. Leaders don't remember data — they remember clarity. The fix is structural: start with where you are now, then what informed it, then what happens next.
End every update by naming the role you need leadership to play:
Formula
Here's where we are.
Here's what informed that.
Here's what happens next.
Example
"We're moving forward with vendor A.
They beat the others on speed and cost by 30%.
Contract goes out Friday."
Blocker Format
"We've hit a blocker with X. I see two options — A or B. I'm leaning toward A because of Y. Need your input?"
Infrastructure
The competitive shift is from "who has the strongest model?" to "who runs the fastest, cheapest, most reliable feedback loop around the model?" The harness team may become as important as the model team.
Winning Stack
No AI-native coding agent company is crushing competitors by being better with AI alone. The competitive stack is now model plus harness plus evaluation loop — not merely a stronger base model.
The Harness Team
DeepSeek is building a dedicated harness team to close the loop between model outputs, runtime feedback, validation, and correction. The harness is becoming a first-class engineering discipline.
Compounding Loops
A cached-input cost advantage makes tighter interaction and verification loops more economically practical. More correction passes at lower cost compounds into a structural reliability advantage.
References
Pragmatic Programmer discussion with Dax Raad (OpenCode)
Leadership update structure — Instagram reel
Zhihu / DeepSeek harness summary
Design Notes
YIL 2026-05-27 — AI Made Software Cheaper to Type, Not Cheaper to Think About
A day examining how AI shifts the economics of software work — reducing implementation cost without removing judgment — while the real competitive moat moves from model strength to harness quality, and leadership clarity turns out to be the same feedback-loop thinking applied to communication.
This page ports Cohere's sober enterprise AI command-center aesthetic — white canvas, monumental tight typography, and deep enterprise-green product bands — to a day about the shifting economics of AI-assisted software work. The palette maps naturally: coral for the editorial/economics discussion (Cohere uses it for blog taxonomy and warm markers), action blue for the clarity-and-communication section (Cohere's editorial link color), and deep enterprise green as a full-bleed product band for the harness-engineering section (Cohere's North/Command product environment color). The page opens with a single oversized typographic declaration on white — no imagery, no decoration — matching Cohere's home-page conviction that size, tracking, and empty space do all the work. The harness section gets the most visual weight: a full-bleed deep-green band styled like Cohere's product capability environments, reinforcing that this isn't a casual observation but an infrastructure-level strategic shift.