The Agentic Harness: How We Rebuilt Our SDLC for AI-Augmented Product Engineering
Prakash Donga · 8 October 2026
Read the originalPrakash Donga of SoluteLabs describes a four-phase process for AI-written code, with a human check after each phase and a second AI agent that reviews before a person does.
Prakash Donga is chief technology officer of SoluteLabs, a firm that builds software for clients. A three-line configuration change broke the navigation layout of its site, and the fix was known to only one engineer. The firm then rebuilt its delivery process so that such lessons are written where engineers and AI agents read them before changing code.
- The process has four phases: gather context, write a plan, build and verify. A person checks the scope, then the architecture, then the code change, then the evidence before release.
- Constraints learned from real failures are recorded in CLAUDE.md, a file in the repository that engineers and AI agents read at the start of a session.
- For larger changes, one AI agent writes the code and a separate review pass checks it before a person sees it.
- Donga names the risk as verification debt: AI produces code faster than the team can prove it is safe to ship.
An engineering team can write down who checks what at each stage of AI-assisted work before it speeds up the building. This is an inference; the piece is SoluteLabs' account of its own process, gives no measurements and ends by offering its services.
When a production failure exposes a rule the code does not show, write it in the instruction file that engineers and AI coding agents read before changing code.
Have a separate AI review pass check each larger AI-written change before a human reviewer sees it.
Derived by Working Surface from the article; more in the Playbook. Source line: AI can accelerate execution without taking ownership of accountability.
9 October 2026: when AI agents take on routine work, teams write down in advance what a person still decides, put those rules where the agent reads them, and keep the people who approve able to understand the work.