Software Engineering Daily — "Organizational Context for AI Coding Agents" with Dennis Pilarinos

Date: March 5, 2026
Host: Kevin Ball "K. Ball"
Guest: Dennis Pilarinos, Founder & CEO of Unblocked

Episode Overview

This episode dives deep into the shifting challenges of AI-driven software development, focusing on what it means to engineer and maintain context in organizations harnessing AI coding agents. Dennis Pilarinos shares insights from his decades building developer tools, discussing Unblocked’s mission to address the “context gap” — the challenge of reconciling scattered and inconsistent organizational knowledge both for humans and AI agents. Discussion spans context engineering, permission-aware AI, organizational adoption, and the changing nature of software work in the age of intelligent agents.

Key Topics & Insights

1. The Rising Importance of Organizational Context

[00:00–06:09]

• Problem Shift: As AI agents handle more code generation, bottlenecks are moving from “writing code” to “understanding the why”: architectural tradeoffs, historical decisions, and the broader organizational context.
  • “The hardest problems are shifting away from code generation towards something context.” (A, 00:00)
• Unblocked’s Solution: Aggregates knowledge across source code, PRs, documentation, chats, and telemetry — serving as a “context engine” both for humans and for AI-driven agents.
• Dennis’ Background: Two decades working on developer tools: from Microsoft Azure to AWS, founding BuddyBuild (mobile CI), then Apple.

  "Unblocked helps engineers get the context they need to get their jobs done." (C, 02:17)

2. What Do We Mean by 'Context'?

[02:59–05:34]

• Context For Machines: All information required for "decision-grade" outcomes.
• Context For Humans: Includes tribal knowledge, why things were done, who to talk to, history of decisions.

  “Even setting context about context feels very meta.” (C, 03:13)

• Where Context Lives: Scattered across Slack, Teams, code history, PRs, bug trackers, documentation, production telemetry — not unique to AI, but now impacting agents and people alike.

3. Reconciling Sources of Truth & The “Context Engine”

[06:09–07:44]

• The Search Problem: Finding scattered knowledge, and reconciling contradictions across tools (e.g., code says one thing, Slack discussion another, Jira a third).
• Conflict Resolution: The goal is to behave like the most knowledgeable team member who’s “been in every conversation.” Unblocked aims to provide answers with high trustworthiness.

  “Imagine talking to someone who just constantly lied to you... you’re not going to trust them.” (C, 06:09)

4. Developer & Agent Experience with Unblocked

[07:44–09:49]

• For Developers: Started as Q&A—ask questions in a Mac app or web interface, with powerful integrations into Slack/Teams channels.
• Multilingual Support: Enables cross-geo, cross-language teams to ask in their preferred language; avoids “judgment” for non-native speakers.
• For Agents: Agents now have access to the same context as people — when they write code, they benefit from historical, architectural, and workflow knowledge.

  “As part of their reasoning experience, [agentic tools] have the full end-to-end context to make the decision for when they're actually ... writing code.” (C, 09:33)

5. Integrations & APIs

[09:49–10:40]

• Connecting Data Sources: Hook up Slack, Notion, code repos, etc., for unified knowledge extraction.
• Flexible Interfaces: Exposed as APIs, command-line tools, “skills,” and even (jokingly) WSDL for legacy fans.

  “We're recording this on a Monday, I think by Friday this will change once again.” (C, 10:13)

6. Handling Discrepancies and Temporal Queries

[10:40–13:34]

• Types of Questions: "How does it work now?" vs. "How did we get here?" — temporal aspects matter for both agents and humans.
• Temporal Reasoning: System weights and analyzes input sources by time and identity.

  “LLMs tend to be bad about time ... how do you think about that?” (B, 11:59)
  “You basically have to take all the artifacts and weight them accordingly and come up with what you believe to be the truth at that point.” (C, 12:23)

7. Permission-Aware Context & Access Control

[13:34–15:26]

• Granular Permissions: Answers are only constructed from data the user (or agent) currently has access to; runtime enforcement prevents privilege escalation.

  “If you don't have access to the underlying data source, the context engine doesn't use it as part of its response.” (C, 13:55)

• Demo Example: If access to a Google Doc is revoked, Unblocked ceases to answer questions involving it.

8. Context Engine Architecture & Data Challenges

[15:26–18:35]

• No Well-Organized Organizations: Most docs are out-of-date; expecting teams to maintain perfect docs is futile.
• Automated Data Enrichment: Ingests and analyzes all PRs, computes diffs, and builds a knowledge graph—even when textual context is sparse.
• Identity Binding: Cross-correlates identities (GitHub, Slack, Notion) for permission consistency.
• Hybrid RAG with Agents: Utilizes retrieval-augmented generation plus agentic orchestration, bridging search and autonomous reasoning.

  “If the tool can call into that API... you can start to really light up a whole bunch of these scenarios.” (C, 17:32)

9. Observability & Debuggability of Agentic Systems

[18:35–21:22]

• Traceability: Instrumentation lets users see if Unblocked was “helpful,” and to probe explanations for agent actions.
• Quality-vs-Speed: Users overwhelmingly prefer slower, more thorough (accurate) results versus fast, rough answers.

  “It's an overwhelming majority of people who are willing to wait for a higher quality answer.” (C, 20:32)

10. Human-Agent Workflows and Mental Models

[21:22–27:18]

• Plan Generation: Increasingly, agentic workflows rely on rich plan generation that incorporates context and avoids pull request bottlenecks.
• Updating Organizational Knowledge: Q&A logs and context traces help humans understand what agents did — maintaining alignment across human and machine mental models.

11. Keeping Context Current & Managing Drift

[23:31–27:18]

• Automated Updates: Agents may perform (or suggest) updates to tickets, docs, or chats, depending on organizational comfort levels.
• Drift Handling: Source code is treated as the ultimate source of truth; indirect context is inferred through observing artifact evolution.

  “At the end of the day, source is the truth, right. Of what's actually happening at that point in time.” (C, 24:43)

12. Culture, Documentation, and the Impact of Remote Work

[27:18–29:22]

• Cultural Predisposition: Organizations’ habits—remote vs. in-person, degree of documentation—affect the quality of usable context.
• Pandemic Impact: Forced remote collaboration led to more being written down, fortuitously improving AI and agent readiness.

13. Legacy vs Greenfield — Divergent Productivity

[29:22–30:44]

• Greenfield Projects: Don't need as much context, AI productivity feels almost magical.
• Legacy Codebases: Context requirements are huge; lack of documentation or clarity increases agentic struggles; Unblocked reduces "REPL time" for agents here.

14. Code Review Bottlenecks in the Agentic Era

[30:44–34:08]

• The New Bottleneck: Human review struggles to keep pace with AI-generated code deluge.

• Unblocked’s Code Review Tool: Uses the context engine to enforce best practices and standards, natively catching issues that AI-written code might introduce.

  “There's been more technical debt created at big banks in the last six months than there has been in their entire history of existence.” (C, 31:16)

• REPL Loop Shortening: AI can auto-remediate code review feedback, raising questions about the future necessity of traditional PR workflows.

15. The Changing Role of Software Engineers

[34:08–38:15]

• Evolving Identity: As agents write more code, engineers’ focus shifts further toward architecture, intent, and business alignment.
• Framer Analogy:

  “Just because you have the tools to build that scaffolding faster ... doesn’t mean you don’t have the craftsmanship.” (C, 36:54)

• Ultimate Metric: The business problem being solved—door “swings the right way”—is what matters.

16. Extending Context Beyond Software Engineering

[38:44–41:07]

• Non-Dev Use Cases: Adjacent teams (product support, solutions engineering) leverage Unblocked for similar context-heavy problems.
• Writing Style Analysis Example: Used Unblocked “context engine” to compare and adapt writing style by mining Slack, Notion, and blogs — not just code.

17. The Unblocked Roadmap & SDK/API Vision

[41:07–44:37]

• Evolving Platform: From Q&A, to context engine, to code review product, now toward exposing engine as SDK/API, so companies can “context enable” their own workflows and agents.

  “Organizations are building their own tools, their own applications that need that context engine ... increasingly more and more people are going to build apps within their enterprise and Unblocked can play a significant role.” (C, 42:47)

• Integration Levers: Users can tune context priority, define sources of truth, and expose custom knowledge sets.

18. What Hasn’t Changed in Software?

[44:37–47:28]

• BXT Model: Always begin with business problem, then experience, then technology.

  “Almost all the projects ... are as a consequence of my frustration with the inability to do something…” (C, 45:00)

• Human Bottlenecks Endure: The “guy who knows code signing at Apple” — domain experts become the bottlenecks; Unblocked aims to reduce dependency on “bothering” these people.

19. The Frontier: Unsolved Problems in Context Engineering

[35:32–36:49; 47:47–48:15]

• Open Questions: Keeping engineers up-to-date as agentic workflows take over; supporting on-call/DevOps for systems written by agents; the evolving definition of “software engineer.”
• Craftsmanship & Differentiation: In a world of rapid, agent-driven software generation, quality and business alignment will set winners apart.

  “While apps are being built very rapidly, which ones are going to be commercially successful ... is the ones that have great craftsmanship.” (C, 48:01)

Notable Quotes & Key Moments

• On Context's Centrality:

  “We'd like to be thought of as like a member of your team who's been aware of every conversation that's ever been had for every part of your code base.” (C, 06:47)

• On Temporal Reasoning & LLM Weaknesses:

  “LLMs tend to be bad about time ... you basically have to take all the artifacts and weight them accordingly and come up with what you believe to be the truth at that point.” (C, 12:23)

• On Agent/Engineer Mental Models:

  "How do we as engineers update our brains on what the agents are doing?" (B, 21:22)

• On Permissions & Security:

  “At runtime you have to be able to understand who that person is. So a strong sense of identity, what access control they have …” (C, 14:15)

• On Rate of Change:

  "It's remarkable to see how advanced some people and organizations are ... and how some folks are still reticent or potentially confused or concerned..." (C, 48:24)

• On What Remains Constant:

  “At the end of the day, the software is a means to an end to solve a specific problem for someone.” (C, 44:57)

Timestamps of Critical Segments

• Dennis’ Background & Unblocked's Mission: [02:07–02:59]
• What Is Context?: [03:13–05:34]
• The Context Search and Truth Reconciliation Problem: [06:09–07:44]
• Unblocked's User Experience & Agent Integration: [07:44–09:37]
• Temporal Reasoning in Context Engines: [11:59–13:34]
• Permission/Access Control Demo: [13:55–15:05]
• Data Layer & Knowledge Graph Building: [15:26–18:35]
• Observability & User Feedback Trends: [19:20–21:22]
• The Greenfield vs Legacy Divide: [29:39–30:44]
• Code Review Bottleneck and Unblocked's Role: [31:16–34:08]
• Frontier Problems & The Evolving Role: [35:32–36:49; 47:47–48:15]
• Engine as Platform & Integration Vision: [41:25–44:37]

Episode Highlights

• Context is the new bottleneck: As agents increasingly write code, the lack of easily retrievable, consistent organizational knowledge is now the main challenge in software development.
• Unblocked's approach: Acting as an omniscient, permission-aware team member, Unblocked provides both people and agents with unified, trustworthy, and actionable context — reducing interruptions, accelerating onboarding, and powering code review.
• Engineering for drift and temporal logic: Historical and temporal reasoning, source-of-truth reconciliation, and fine-grained access control are engineered into the product.
• The future is platform-driven: Unblocked is evolving to offer its context engine as an SDK/API, enabling organizations to context-enable their own tools and agentic workflows.
• Role of the engineer is changing: As agentic development accelerates, engineers must shift focus towards architectural and business alignment, quality, and ensuring the "doors swing the right way."

End of Summary