AI Summary
→ WHAT IT COVERS Dennis Pilarinos, founder of Unblocked, explains how the primary bottleneck in AI-assisted software development has shifted from code generation to context — specifically, aggregating organizational knowledge from Slack, pull requests, Notion, source code, and production telemetry to give both developers and coding agents decision-grade information. → KEY INSIGHTS - **Context conflict resolution:** When Slack, Jira tickets, and source code contradict each other, treat the source code as the highest-weight truth signal, then use pull request diffs as secondary breadcrumbs. Unblocked's engine performs runtime conflict resolution across all connected data sources rather than relying on static, pre-indexed snapshots. - **Permission-aware AI systems:** AI context engines must enforce access control at query runtime, not at indexing time. Unblocked demonstrates this with a live demo: revoking a user's Google Doc access causes the system to immediately stop surfacing that document's contents, preventing privilege escalation through AI intermediaries. - **Pull request enrichment:** Most developers write minimal PR descriptions, leaving organizational knowledge underutilized. Unblocked automatically diffs every pull request in connected repositories and builds a background knowledge graph from those diffs, upgrading low-quality documentation into a queryable institutional memory without requiring any manual effort from engineering teams. - **Planning before agentic execution:** Teams using Unblocked's context engine during the planning phase — before agents write any code — produce significantly better implementation plans because the agent already knows coding standards, architectural decisions, and common code review failure points, eliminating rework cycles that would otherwise surface during pull request review. - **Slack integration as primary adoption vector:** Unblocked monitors engineering support channels and responds to developer questions within seconds when confidence is high. Across tens of thousands of user feedback responses, 95% report the tool as helpful, with roughly half of the remaining 5% simply refusing to answer surveys rather than indicating dissatisfaction. → NOTABLE MOMENT Pilarinos tested his own product by letting Claude Code handle secrets management without context — it chose a method that would expose credentials immediately upon infrastructure compromise. Unblocked's code review caught the error and recommended vault-based encryption, which Pilarinos then delegated back to the agent to fix autonomously. 💼 SPONSORS None detected 🏷️ Context Engineering, AI Coding Agents, Developer Tooling, Organizational Knowledge Management, Agentic Workflows