Two years ago we argued that when agents write most of the code, the bottleneck shifts from “can we build it?” to “do we want it? is it right?” Product and tech could finally focus on craft—strategy, judgment, coherence, learning—instead of queue management. Less queue, more craft.

How did that thesis hold up?

What Held Up

The bottleneck did shift.
Teams using agentic workflows report that implementation is less the constraint than it used to be. The scarce resource is now deciding what to build, validating that it’s right, and ensuring it coheres. That part of the thesis was correct.

Craft became more visible.
When boilerplate and routine implementation are handled by agents, the work that remains is more obviously “craft”—architecture decisions, product strategy, user research interpretation. Teams that leaned in did experience a shift toward higher-leverage work.

The narrative mattered.
”Queue to craft” resonated. Teams that adopted that framing had an easier time with desire and reinforcement. The alternative framing—“AI is coming for your job”—triggered resistance. How we told the story made a difference.

What Shifted

The bridge was harder than we thought.
Getting from prototype to production stayed difficult. Agent output is getting better, but production readiness—consistency, edge cases, maintainability—still requires significant human work. The “less queue” part was real; the “more craft” part often included “glue work” we didn’t fully anticipate. Craft, but also integration and validation.

Tiny teams need different designs.
We celebrated the possibility of tiny teams running big systems. We understated the design work required: observability, failure modes, and the need for generalists with depth. Tiny teams can punch above their weight—but they need systems designed for that reality.

Reinforcement is the hard part.
Awareness, desire, knowledge, ability—teams made progress. Reinforcement, consistently, was where adoption plateaued. Making the new way stick required more than we initially prescribed: defaults, constraints, and sometimes sunsetting the old way. That’s politically and operationally hard. We could have emphasized it earlier.

What We’d Do Differently

Lead with the bridge.
We’d put “prototype to production” front and center. It’s the gap that determines whether agentic workflows deliver value. Invest in it from day one.

Design for Reinforcement from the start.
Don’t assume adoption will sustain itself. Build defaults, metrics, and reinforcement into the initial design. Treat it as a first-class concern, not an afterthought.

Be more explicit about glue work.
”More craft” is true—but it’s also “more integration, more validation, more orchestration.” Some of that is craft; some is operational. Being clearer would have set better expectations.

The Verdict

Less queue, more craft: still right as a North Star. The path is messier, the bridge is longer, and Reinforcement is harder than we expected. But the teams that committed to it are playing a different game. And that was the goal all along.