Dennis Pilarinos from Unblocked: context for AI coding agents

The discussion centers on the importance of context in enabling both developers and AI agents to effectively solve problems. Initially designed as a context layer for developers, Unblocked has evolved to support agentic workflows by providing AI agents with organizational knowledge that mimics what a new developer would need on their first day. Without proper context, agents struggle with inefficient tool usage, redundant searches, and poor decision-making, leading organizations to manually feed them information. Challenges include managing outdated or contradictory data, scaling context maintenance, and integrating unwritten or tacit knowledge into usable systems.

To address these issues, the conversation highlights the need for a unified context layer - such as a knowledge graph - that aggregates, reconciles, and dynamically updates information from disparate sources like Slack, Notion, and Confluence. This approach avoids the limitations of static documentation or multi-source retrieval methods that may lead to incomplete or conflicting answers. The benefits of such a system include faster task completion, reduced token usage, improved decision accuracy, and seamless integration into existing workflows without requiring users to reorganize data. Additionally, identity resolution across platforms and enforcing access controls are emphasized as critical components to ensure secure and accurate AI interactions.

• What if you built a real-time context layer for your solo dev AI agent today?

  • Move: Integrate your AI coding assistant with live data sources (e.g., GitHub, Notion, Slack) using Unblocked or a custom knowledge graph pipeline to dynamically pull in project context before every task.
  • Why Now?: Static READMEs and notes decay quickly; your solo workflow depends on up-to-date tribal knowledge (like recent decisions or API changes) that only lives in fragmented messages or commits. Waiting erodes velocity.
  • Expected Upside: Reduce AI hallucination and redundant queries by 50%, cut debugging time by at least 30%, and accelerate feature delivery without hiring or documentation sprints.

• What if you offloaded context management from your brain to an automated system this week?

  • Move: Replace manual note-taking and mental recall with a lightweight context sync tool that observes your work patterns (e.g., file edits, PR comments, chat snippets) and auto-generates a searchable, reconciled knowledge base.
  • Why Now?: As a solo operator, cognitive load is your biggest bottleneck - every context switch costs minutes. Tools now exist to capture unwritten logic without forcing workflow changes.
  • Expected Upside: Reclaim 2 - 4 hours per week spent reorienting after breaks, improve AI-generated output accuracy, and create transferable institutional memory for future scaling.

• What if you enforced permission-aware AI access to your private repos and docs tomorrow?

  • Move: Configure your AI tools to respect file-level permissions (e.g., via GitHub OAuth scopes or Unblocked-style identity resolution), ensuring suggestions never expose restricted code or sensitive data - even when pulling from aggregated sources.
  • Why Now?: Solo devs often reuse personal agents across projects; one misstep could leak credentials or proprietary logic. With agentic workflows rising, runtime access control is no longer optional.
  • Expected Upside: Safely leverage AI across all projects - including private ones - without risk, enabling faster iteration while maintaining compliance and IP protection.

• Implement a unified context layer in your software to aggregate data from multiple sources (e.g., GitHub, Slack, Notion) into a reconciled knowledge graph, reducing agent errors and token usage.
• Design AI workflows that respect existing user permissions across systems by dynamically checking access rights at runtime to prevent unauthorized data exposure.
• Avoid requiring users to reorganize or clean up their data; instead, build tools that automatically connect to and enrich existing data sources without workflow disruption.
• Optimize AI agent performance by reducing redundant tool calls and inefficient searches through pre-loaded, up-to-date organizational context - mimicking onboarding for a new developer.
• Measure and leverage quantifiable gains (e.g., 30 - 50% lower token use, 40 - 60% faster task completion) when using managed context solutions to justify investment over DIY approaches like static wikis.