This is my working lab. I build real tools, use them in daily professional work, and record what changes in a public knowledge graph. AIBO, CognitiveOS, and the Guide Mode agent are shipping software; the graph is the ongoing research behind them. Each tool is a thinking aid first — not a replacement for human judgment.
These are not mockups. Each one was built to solve a real friction in AI-assisted work — writing, preparing for hard conversations, reasoning over live speech, or keeping track of what I'm learning — and each one feeds observations back into the research graph.
A Windows tray tool that rewrites, fixes, summarizes, or translates selected text almost anywhere via a hotkey — bringing AI to where the text already is instead of a separate chat window.
A graph-centered mobile workspace for preparing, rehearsing, and supporting high-stakes conversations. A project starts as a thinking graph, becomes a narrative path, and can generate scripts or live guidance.
A Vercel-hosted agent service behind CognitiveOS Guide Mode. It keeps agent instructions, output schema, and API keys off the device, and returns validated graph updates from a live transcript window.
Inside CognitiveOS, Guide Mode listens to a live conversation and quietly keeps you oriented — focusing existing graph nodes, suggesting what to say next, and proposing new nodes from what was just said.
A public knowledge graph of the research itself — 51 concept nodes and 160 relationships across thinking, workflow, decisions, psychology, and the future of work, explorable directly in the browser.
The routine that keeps the graph alive: new notes and observations are merged into the knowledge graph through a structured Claude workflow, turning scattered daily AI use into a connected, revisable record.
The collection is intentionally practical. Tools are judged by whether they improve the quality of thinking, reduce unnecessary friction, and preserve accountability when AI enters the workflow.
Define the thinking job first: briefing, deciding, reflecting, synthesizing, or communicating.
Make inputs, uncertainties, constraints, and interpretive leaps visible before action is taken.
Use AI to organize and challenge thinking while keeping ownership with the person using the tool.
Check privacy, security, traceability, and whether the output can be explained in plain language.
Building and using these tools is the method, not the point. The point is a long-running, self-directed inquiry into what actually changes when AI is deeply integrated into daily thinking and work. These are the questions I'm holding open — and the conclusions I'm arriving at so far.
The feeling of having thought deeply is not proof that you did. AI can produce the shape of insight without the process that makes it stick.
The blank page was uncomfortable — but that discomfort was where original thinking happened. Removing it removes something worth preserving.
Receiving an explanation is not the same as building a model. Real understanding requires integrating new information with everything you already know.
As generation becomes cheap, the bottleneck shifts to framing. How you define a problem sets the ceiling of any AI-assisted work.
Speed and depth are not the same. AI makes execution faster while depth of thinking still requires deliberate human judgment to protect.
The ceiling of individual potential is rising. The constraint is no longer time or access — it is clarity of intent, quality of judgment, and willingness to attempt something meaningful.