The way to build and run software products has changed. At Integral we’re looking for the best way that fits our needs given our smart, small team. That includes building and releasing software, but it doesn’t stop there. We’re slowly documenting our experience: the simple approach we’re taking, avoiding tool-mania, discussing what’s working, and what we’re learning along the way.
The picture below is a snapshot from our whiteboard: our current view of the flow. It’s messy because we’re still shaping it. (and also I’m the one who draws it)
What you’ll find here
Posts in the SDLC2 series describe how we run this in practice:
- Agents as Code: the abstraction layer has moved on, did you? How we treat agent commands and skills like code: one repo, changes via PR, and writing them in plain English together.
- Research, Plan, Implement: How we use agent-resources commands to go from unknowns to shipped code—research-codebase, generate-plan, implement-plan—and how the artifact chain links them.
- Parallel plans, parallel agents (stay tuned): How a single research document can feed multiple plan variants, and how implementation itself can be parallelized across agents.
- Reducing the slob (stay tuned): All these tools and artifacts generate noise. How we keep the organizational footprint of AI-assisted development from becoming a mess in its own right.
- Issues as conversations (stay tuned): How we use agents to draft, refine, and review Linear issues before any code is written — and why the quality of the issue shapes everything downstream.
- Zero-touch shipping (stay tuned): For trivial issues: small fixes, copy changes, config updates — we skip the whole cycle and ship straight to production without a human in the loop.
We’ll add more as we go. If you’re curious how a small team is rethinking the software development lifecycle without waiting for the industry to settle (how we learned to stop worrying and love the release, Strangelove-style), this is where we’re writing it down.