I recently sat down with Ashley Smith, Founder and GP at Vermilion Cliffs Ventures, to talk about what it actually takes to build software in the age of AI agents.
Not “how to write code faster.” Most teams have already felt that jump.
The harder question is what happens after the code shows up: when output scales, the bottlenecks move to review, environments, CI, release flow, and operational trust. That’s where teams either compound their advantage, or stall out.
Why this conversation matters
AI tools are accelerating. The constraint isn’t.
When every engineer can generate more code—and can run multiple agents in parallel, the old engineering workflow starts to break:
- PRs pile up faster than teams can review them
- shared environments become contention points
- CI becomes the gate that everything queues behind
- release processes become the limiting factor, not implementation
That’s the central idea we explored:
- The tools will keep improving quickly
- Workflow design is now the real constraint
- Winning teams will intentionally decide where humans stay in the loop—and where agents can run
A few themes we covered
From individual productivity to team throughput
“Faster coding” is only the first step. The real leverage is shifting from developer speed to system throughput: how quickly an organization can take change from idea → verified → shipped.
Parallel engineering with agents
When each engineer can run several agents at once, you don’t just get more output, you get different failure modes (coordination, integration, review load, and quality drift) unless the workflow is redesigned around parallelism.
Human‑in‑the‑loop by design
The goal isn’t to remove humans, it’s to put judgment in the places where it’s highest value: product intent, architecture, risk, and correctness. Everything else should be engineered so it can scale.
Infrastructure as leverage
Environments, previews, CI, and release flow matter as much as model quality, because they determine whether increased output becomes shipped value or permanent backlog.
What I believe (and where Wallfacer fits)
We’re in a platform shift. The biggest advantage won’t come from sprinkling AI tools on top of yesterday’s process.
It will come from teams that rebuild their engineering system to match what agents can do now, and what they’ll be able to do next.
That’s why we’re building Wallfacer: to help teams close the gap between what agents can produce and what organizations can reliably ship.
If you’re an engineering leader navigating this transition, I think you’ll find this episode useful.
And if you’re actively redesigning your workflow for agent-era engineering, I’d like to hear what’s breaking for you and what you’re experimenting with.
Thanks again to Vermilion Cliffs for having me on.