Our Thinking
Perspectives on AI, expertise, and why the hard problems matter most.
May 28, 2026 · William VanBuskirk
Joining NVIDIA Inception, and moving down the stack
We're now a member of the NVIDIA Inception program, and we're using it to move down the stack: from steering models with context toward tuning them.
May 13, 2026 · William VanBuskirk & Mark Freedman
Hello, world. This is myai.
After the better part of a year building mostly with industrial customers, the public version of myai is here. Watch the trailer, see where to start, and meet the platform that meets you where you are.
May 4, 2026 · William VanBuskirk
Most AI stacks are built backwards. Start with the agent, not the data.
The default enterprise AI sequence (data, CRUD, dashboard, agent-on-top) is the wrong shape for agents. The reframe is the audience: the agent isn't the last reader of your data, it's the first.
May 4, 2026 · William VanBuskirk
Use cases ship. Operating models work.
The industry sells agent use cases (meeting summarizer, support copilot, sales-research bot). The unit of design that actually scales is the operating model: an agent with a beat, peers, escalation paths, and artifacts it owns. The agent goes on the org chart, not in the feature list.
May 2, 2026 · William VanBuskirk
Voice agents don't need a faster horse. They need a dispatcher.
Why every consumer voice agent feels weirdly limited the moment you try to do real work in it, and how the same architecture that makes Claude-via-MCP feel deep makes voice-to-voice possible without breaking the loop.
April 27, 2026 · William VanBuskirk
Sparse but Deep: why most RAG is noise
Last week we ran the same question through two retrieval shapes: Codex over a week of commits, myai over the artifact graph. Same input, very different outputs. Here's why we summarize before we search, and why it matters more the longer your agent runs.
April 26, 2026 · William VanBuskirk
myai is the context. Your agent is the hands.
Why we deliberately don't try to be a catch-all MCP connector. What becomes possible when myai handles your context and your agent of choice does the rest.
April 13, 2026 · Mark Freedman & William VanBuskirk
Your agent instructions are describing, not prescribing
Almost all agent documentation (system prompts, CLAUDE.md files, cursor rules) describes the world. The document your agent needs is the one it can't derive from the code: your intent.
April 12, 2026 · William VanBuskirk
Your workarounds are telling you something.
The enterprise software industry has spent 30 years telling people their workarounds are wrong. What if the workarounds are the smartest thing in the building?
April 9, 2026 · William VanBuskirk
Why Your AI Agent Needs Dimensions, Not Just Skills
Most AI tools organize knowledge in a big bucket. But no one's actual job looks like that. Jobs are multi-dimensional, and your agent architecture should be too.
March 28, 2026 · Mark Freedman & William VanBuskirk
The Socratic Superpower: Why Asking AI Questions Matters More Than Prompting
Most people interact with AI by screaming at it. The real leverage is in the back-and-forth.
March 25, 2026 · Mark Freedman & William VanBuskirk
Beyond easy problems: where AI actually creates value
Everyone's automating outbound emails and meeting notes. The real value is in coordination problems that cross manufacturing, engineering, and supply chain.
Bring us the mess.
We’ll show you the map.
Every conversation starts with your problem, not our pitch. Book 30 minutes and we’ll explore whether myai fits.