Brandon Waselnuk

Brandon Waselnuk is a Canadian startup founder, developer advocate at Unblocked, and venture partner at Pioneer Fund. He develops approaches to organizational context for coding agents: giving automated software developers the institutional knowledge they need to produce work an engineering team can actually merge.

From founder to developer-tools advocate

Waselnuk worked in product strategy and design thinking at IBM, led product strategy at Assent Compliance, and worked in venture capital at CoVenture. In Ottawa, he co-founded and led Tattoo Hero, which built software for tattoo shops and artists.

He subsequently founded Codex with Karl Clement and Saumil Patel. Part of Y Combinator’s Summer 2021 batch, the company created a Visual Studio Code extension that attached questions and shared engineering knowledge directly to relevant code. It later operated as Squire AI, building tools for pull-request descriptions, code review, and implementing review feedback. Waselnuk’s Codex was an independent startup, distinct from OpenAI’s similarly named product.

In November 2024, he joined Mintlify as head of product, having previously used its documentation platform as a customer. His account of joining Mintlify described documentation as essential infrastructure for both human understanding and AI. He later joined Unblocked, where his work addresses context across repositories, pull requests, architecture decisions, documentation, and workplace conversations. In October 2025, he announced his return from San Francisco to Vancouver.

What coding agents actually need

• A context engine, not a collection of connections. Access to workplace systems through the Model Context Protocol cannot tell an agent which information matters. Waselnuk advocates dynamically assembled, concise research packets that combine organizational knowledge before an agent plans or writes code.

• Mergeable code depends on institutional knowledge. Software can compile and pass checks while duplicating an internal service or violating established architecture. In a Zendesk integration demonstration, contextual retrieval identified existing provider-registration mechanisms and factory patterns before the agent explored the codebase.

• Permission-aware social graphs make retrieval relevant. Contribution histories, review relationships, repository ownership, and technical expertise help determine what a request means for a particular engineer. Waselnuk describes keeping private conversations inaccessible to unauthorized users as a design requirement. The retained documentation commentary states that Slack permission enforcement is always enabled, other supported sources require configuration, and some sources do not support it.

• Conflicting and stale evidence requires judgment. Waselnuk warns against satisfaction of search: accepting the first plausible result while missing contradictory discussions or better-established patterns. Cached answers can also become misleading as code and decisions change. His writing on AI-assisted code review similarly favors useful architectural feedback over obvious comments that consume engineers’ attention.