May 1st was my last day at Augment Code. Jotting down some reflections.

The topline: Grateful for the kind and supportive folks I met here, especially the design team: Matt, Robbie, Todd. I’m leaving mostly because I’m inspired by what I’ve learned, and I am looking for space to explore ideas.

Also: Sometimes, building before alignment is simply wasting tokens. More on this later.

Gratitude

At Augment, I got to work on two projects that were very meaningful to me. First is Augment Intent. Getting to work with Amelia is amazing and an honor. For months, it was mostly the two of us going out on a limb to explore what agentic software development should look like. And we really got to explore, “hey, this technology is changing. It’s changing some fundamental assumptions. How is this work going to be done?” There’s no consensus, no template. Indeed, the whole point was to see if we could break the mold. Hard to imagine a more fun project than that.

The second one is this code map visualization, which unfortunately we never got to ship. When I first joined Augment, one of the first things that my manager Matt and I talked about was visualizing the context engine. Its ability to index the codebase and find relevant code to inform the agent is the technical differentiator for Augment. It was a big part of why I joined them. A core piece of the context engine is the code embeddings, and I had a hunch this could be used in a cool visualization.

So I tinkered with it for months. Prototyping different layouts, learning about performant visualizations in threejs … some pretty standard front-end stuff. More importantly though, I delved into various kinds of data science and data pipeline work to transform and enrich the embeddings data to produce an engaging map. It was this work that, without using a coding agent, would have been beyond my ability – this was the work that was the most meaningful and transformative. I had to learn and lean into using coding agents, and there were not many places other than Augment where I could have done this. I came out of this with a radically changed workflow, and a transformed sense of what I am capable of.

… but it didn’t ship!?

Unfortunately, no, it didn’t. It simply didn’t align with what the company is focused on anymore. Augment is pivoting towards an agentic automation platform. A code map doesn’t fit into that story. And I was too caught up building the thing, and didn’t do enough evangelism and strategy work. In the spirit of lemons into lemonade …

Not shipping was an interesting lesson. Lots of folks online talk about how this is a golden age for building. Or how coding agents are deluding designers/engineers/product managers into thinking they no longer need “the other two legs of the stool”. I think I fell into that trap.

Coding agents change the dynamic of how teams work together. In this maps project, I built something that is shippable! No support from product or engineering required! I have a repeatable pipeline that can ingest codebases, and spit out the visualization! It is literally on a preview branch on Vercel right now. Augment can literally flip a switch and it will be publishable right now.

But it didn’t ship. Why?

What I didn’t do is the institutional work of getting alignment, and bringing people along the journey. There are many complex reasons why things don’t ship — but I know that at least for my part, I need to do better at alignment. Sometimes, building before alignment is simply wasting tokens.

Alignment is the Bottleneck Now

I suspect this failure mode will be common over the next couple years as the industry adapts. Sure, you can build more things now, but you don’t necessarily have the institutional buy-in to go ship stuff. And I think one of the looming things coming up in the ecosystem is: all this stuff will get built. How will they get shipped? How do they get metabolized in the product organization?

Or, will product organizations just fall apart? Maybe in the next five years, we devolve even further into a VC model. Big companies no longer even try to do meaningful product development internally. Just buy the team that built the thing they wanted, and integrate? Crazy thought.

What’s Next

Nothing! Well, not exactly nothing. I’m not taking another job, and not trying to build products. Instead, I am going to give myself six months to do some “directed play.” I have long wanted to try my hand at publishing more weird stuff, like R2D3. With all this experience with coding agents (and now also a green card) there’s no better time? So I’m going to do the kind of work that I personally want to do. Then publish, publish, publish, as much as I can. In six months, let’s see if I can get anybody to be interested in paying me to do this kind of work. A sort of “bet on myself” moment.

Honestly, I am kind of terrified. In six months, who knows if there’s still a product design job market? At the same time … my sense is that the only way out is through. It would be better to transform myself now to prepare for whatever replaces “product design” as a title.

So, might as well do it.