Why AI agents fail at code: the context gap Unblocked is solving
AI agents are writing more code than ever. The bottleneck has moved. It’s no longer about generating code. It’s about giving agents enough context to make code that actually meets production standards.
That’s the argument Dennis Pilarinos, Founder and CEO of Unblocked, made on a recent Software Engineering Daily episode with Kevin Ball. Pilarinos has the background to back the claim: he helped build Azure at Microsoft, worked at AWS, and co-founded BuddyBuild, the mobile CI platform Apple acquired.
The Problem: Context Lives Outside the Codebase
According to Pilarinos, AI agents fail when they lack access to the organizational knowledge that explains why a system works the way it does. Architectural decisions, historical pull request discussions, chat threads, production telemetry, and external documentation all carry context that code alone cannot surface.
Unblocked’s context engine aggregates and reasons over knowledge spread across source code, pull requests, documentation, chat systems, and production telemetry. The goal is to serve that context to both human developers and AI agents in real time.
️ What Unblocked Targets
The platform is built around four operator pain points that teams hit as they scale agentic workflows:
- AI code quality and review: Agents produce output that looks correct but misses institutional decisions baked into older parts of the system.
- Developer interruptions: Engineers spend time answering context questions that should be self-serve.
- Onboarding speed: New developers and new agents both need ramp time when organizational knowledge is scattered.
- Agentic workflow safety: Without grounded context, autonomous agents make changes that conflict with existing constraints.
The Conversation
The episode covers context engineering as a discipline, how to reconcile conflicting sources of truth inside an org, permission models for AI systems, the shifting bottlenecks in the software development lifecycle, and what the role of software engineer looks like as agentic tools take on more of the execution layer.
Kevin Ball, who serves as VP of Engineering at Mento and runs the AI inaction discussion group through Latent Space, hosts the discussion. The episode is sponsored by Unblocked.
If you’re building with AI coding agents and finding that output quality degrades as system complexity grows, this is the clearest articulation of why that happens and what a structured fix looks like.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
