Visual summary of operating lessons from Dennis Pilarinos.

Lessons from Dennis Pilarinos

Dennis Pilarinos founded the mobile CI platform Buddybuild, later acquired by Apple, and now leads Unblocked. He argues that coding agents need organizational knowledge—how systems work and why decisions were made—to produce useful work in an existing codebase. This profile covers his lessons on developer tools, company-building and AI context. — Software Engineering Daily Transcript.

Part 1: The Context Gap and AI

  1. Context before code: For AI coding agents, generating code is only part of the challenge. They also need decision-grade knowledge of why an organization built its systems as it did. — Software Engineering Daily Transcript.
  2. Context as a discipline: Context engineering means deliberately selecting the information an agent needs for a task, keeping it current, authoritative and permission-aware instead of adding documents indiscriminately. — Context Engineering.
  3. Measure useful token yield: Pilarinos argues that AI-tool costs should be judged by useful, mergeable output per unit of spend, not token prices or raw token consumption alone. — AI Tokenomics.
  4. Context and reliable AI code: AI-generated code can look plausible while missing the organization’s actual conventions or constraints; supplying relevant context can make the output more reviewable. — Context Engineering.
  5. Resolve conflicting sources: A context system has to reconcile differences among code, tickets and conversations; simply retrieving more material can give an engineer or agent contradictory answers. — Software Engineering Daily Transcript.
  6. Knowledge across tools: The knowledge needed to understand a codebase can live in source history, pull requests, chats, bug trackers, documentation and production systems—not just the repository. — Software Engineering Daily Transcript.
  7. Legacy code needs its history: For AI-assisted legacy refactoring, Pilarinos recommends giving the agent the decisions behind the code and pairing that context with characterization tests, small changes and verification. — Refactoring Legacy Code with AI Agents.
  8. Avoid expensive guesses: When agents act on incomplete or stale context, humans can inherit review and rework. Pilarinos treats context quality as an upstream control on that burden. — Context Engineering.
  9. Context helps new hires too: Pilarinos treats onboarding as context loading: new engineers need to learn why the team works as it does, who to ask and where decisions live. — Software Engineering Daily Transcript.
  10. Reduce repeated interruptions: Unblocked began as a Q&A tool for recurring engineering questions; Pilarinos says answering them in team channels can spare people from waiting on colleagues. — Software Engineering Daily Transcript.

Part 2: Designing Tools for Developers

  1. Build for developers’ real workflow: Pilarinos approaches developer tools from the user’s concrete workflow pain rather than the novelty of a technology; the tool has to make a difficult task meaningfully easier. — First Round Podcast.
  2. Earn developer trust early: Developers have strongly held opinions and little patience for sales language; Pilarinos says a tool has a narrow first-impression window in which to show real utility. — First Round Podcast.
  3. Hard problems can create a moat: Pilarinos is drawn to technically hard problems. He argues that solving one can create defensibility, although difficulty alone is not proof of a viable business. — It Shipped That Way Podcast.
  4. Fit the existing workflow: Buddybuild was designed around a developer pushing code and getting a build on a device, illustrating Pilarinos’s preference for a simpler experience around an existing task. — First Round Podcast.
  5. Mobile delivery had distinct friction: Buddybuild began because producing and deploying an iOS build to a device was unexpectedly difficult; the team focused its first product on that acute mobile CI pain. — First Round Podcast.
  6. Follow changes in the platform: Pilarinos has seen developer ecosystems move from on-premise software to cloud APIs, mobile and AI; each shift changes what tools and business models are useful. — First Round Podcast.
  7. Start from firsthand pain: Pilarinos’s experience with Microsoft and Amazon development systems sharpened his sense that the mobile build workflow he encountered outside them was unnecessarily hard. — First Round Podcast.
  8. Stay curious about rough edges: Pilarinos says he feels a product’s frustrating inefficiencies firsthand; that discomfort helps his team find rough edges and explore a better direction. — It Shipped That Way Podcast.
  9. Look for time returned to users: An unsolicited Buddybuild note about time saved managing build infrastructure was an early signal to Pilarinos that the product mattered to its users. — First Round Podcast.

Part 3: Building a Sustainable Company

  1. Treat founder work as a marathon: Pilarinos says he worked at a sprint pace at Buddybuild and later concluded that distance and sustainable routines make him a better founder. — First Round Review.
  2. Use distance to improve judgment: When Pilarinos steps away from a tangled problem, he says he can return with a clearer forest-level view instead of continuing to hack at one detail. — First Round Review.
  3. Protect founder mental health: After Buddybuild, Pilarinos changed how he works: he protects a day away from work and uses perspective to ride the highs and lows of a new company. — First Round Review.
  4. Expect product-market fit to take work: Pilarinos says even a repeat founder must keep learning from early use, rejection and iteration before a product becomes valuable enough for customers to love. — First Round Podcast.
  5. Make meetings earn their time: Pilarinos questions recurring meetings that solve no current problem, while preserving useful one-on-ones and meetings anchored to a specific decision. — It Shipped That Way Podcast.
  6. Use small-team speed carefully: A small team can build and adjust a product with fewer coordination layers, but Pilarinos notes that a startup must still create the distribution a large company already has. — It Shipped That Way Podcast.
  7. Protect focused work: Pilarinos supported an engineer skipping one of his own meetings to finish important work; the meeting could be resolved without that engineer and the outcome shared later. — It Shipped That Way Podcast.

Part 4: The Realities of Being a Repeat Founder

  1. Experience gives only a rough map: Pilarinos compares founding a second company to moving through a dark room one has visited before: prior experience helps locate obstacles but does not remove them. — First Round Review.
  2. Reassemble trusted collaborators: Ten of Unblocked’s first eleven team members had worked at Buddybuild. Pilarinos says familiarity with one another’s roles and working styles helped the new team move quickly. — BetaKit Interview.
  3. Earn product fit again: Prior founder experience cannot substitute for learning which customers value a new product; Pilarinos describes both Buddybuild and Unblocked as iterative searches for fit. — First Round Podcast.
  4. Adapt after the acquisition: After Apple acquired Buddybuild, Pilarinos moved into Apple’s development-technologies organization. His subsequent company drew on both the startup and large-platform experience. — First Round Podcast.
  5. Starting over changes the founder too: Pilarinos describes over-identifying with Buddybuild, then deliberately changing his work habits and seeking distance from problems before building Unblocked. — Code Story Podcast.

Part 5: Fostering Engineering Culture

  1. Keep product leaders on the support front line: Pilarinos says he and his cofounder answered many Buddybuild support requests themselves, and he continued answering Unblocked users to stay close to the product’s sharp edges. — First Round Podcast.
  2. Scale access to organizational context: As engineering teams grow, knowledge spreads across code, pull requests, conversations and operating systems; Pilarinos argues that new people and agents need access to that history to work effectively. — Software Engineering Daily Transcript.
  3. Give answers, not only documents: Pilarinos argues that documentation is often stale or incomplete; when a developer is blocked, a specific answer that includes decisions and recent discussions may be more useful than another general document. — Developers Don’t Need More Docs.
  4. Treat chat as part of code history: Engineering decisions and workarounds often surface in Slack or Teams rather than code or formal docs, so Pilarinos includes those conversations among the sources needed to understand a system. — Software Engineering Daily Transcript.
  5. Watch for context debt: Pilarinos calls missing, stale or contradictory institutional knowledge context debt: it creates rework when people or agents repeatedly act on an incomplete picture. — Context Engineering.
  6. Find the normalized workflow pain: Pilarinos’s Buddybuild test was whether a frustrating developer task was merely unfamiliar or systemically broken enough to warrant a new product. — It Shipped That Way Podcast.
  7. Make answers available where work happens: Pilarinos describes engineering Q&A in Slack, Teams and the IDE, aiming to let a developer find a useful answer without waiting for a coworker in another time zone. — Software Engineering Daily Transcript.

Part 6: Customer Obsession and Support

  1. Use feedback to calibrate, not replace, judgment: Pilarinos says a team needs its own product thesis and priorities; customer feedback then tests and adjusts that view rather than becoming an unfiltered feature queue. — First Round Podcast.
  2. Apologize while trust can be recovered: With developer users, Pilarinos says a product failure leaves a narrow window to acknowledge the mistake and regain trust; slow support can compound the damage. — First Round Podcast.
  3. Keep builders close to customer frustration: Pilarinos warns that layers between customer feedback and the people building the product can erode empathy; direct support helps builders understand what failed. — It Shipped That Way Podcast.

Part 7: Competing as a Startup

  1. Build distribution alongside the product: Pilarinos says a large company can launch to an existing audience, while a startup starts with almost no distribution and must earn attention as it builds. — It Shipped That Way Podcast.

Part 8: Navigating the AI Transition in Software

  1. Judge AI tools by practical utility: Pilarinos argues that a developer tool should solve a real workflow problem regardless of whether it uses AI; the underlying technology is secondary to what users can accomplish. — First Round Podcast.
  2. Review becomes a constraint as agents write more code: Pilarinos acknowledges that increased AI-generated code can burden maintainers and describes using organizational standards in code review to catch context-dependent mistakes. — Software Engineering Daily Transcript.
  3. Use AI to enhance developer proficiency: Pilarinos framed Unblocked’s early aim as helping developers understand their systems and become more effective, rather than merely typing code for them. — BetaKit Interview.
  4. Treat context as infrastructure: Pilarinos describes a mature context engine as infrastructure that ranks sources, resolves conflicts, checks freshness and permissions, and supplies decision-grade context to agents. — Context Engineering.
  5. Make conflicting evidence visible: Pilarinos says an answer is more trustworthy when it shows disagreements among a runbook, recent pull requests and team discussion instead of silently choosing one source. — Three Hard Lessons from Building Context at Scale.
  6. Modernize legacy code with context and verification: Pilarinos argues AI can speed legacy refactoring only when it has the institutional history behind the code and works inside a guarded loop of characterization tests, small edits and human verification. — Refactoring Legacy Code with AI Agents.