Trailblazer: Capturing Flow States Before They Evaporate

In 2018, I made a video and gave a talk about every band having an AI agent as their tour manager. I designed the concept around the dotblockchain music protocol and the Cougaar Agent Java framework — a DARPA-funded multi-agent system built for military logistics coordination.

It was ahead of its time. Before ChatGPT. Before anyone outside of research labs was talking about AI agents. The core insight was right: touring artists are continuous power sources for legislation, local activism, public awareness campaigns. But they can't manage the logistics, relationships, and accumulated knowledge at scale. An agent could.

That concept evolved into Trailblazer.

The Flow State Problem

The best knowledge workers can't explain their process. Ask them how they found that insight, made that connection, or produced that output, and they'll give you a vague answer. "I just knew." "It came to me." "I followed a hunch."

They're not being evasive. They genuinely can't reconstruct the sequence of actions, context switches, and implicit reasoning that produced the result. Flow states don't leave transcripts. By the time the work is done, the process that created it has evaporated.

This is the knowledge capture problem that templates could never solve. At Rock 'n Renew, we watched brilliant organizers produce incredible programs and then fail to transfer their methodology to successors. The knowledge was in the doing, not in any document.

How Trailblazer Works

Trailblazer uses Playwright browser automation to record your actions while you work. Not keystrokes and mouse movements — semantic actions. "Opened the competitive analysis document." "Switched to the client's website." "Cross-referenced pricing data with the strategy memo." "Searched Slack for the Q3 discussion."

These actions get mapped into a workflow graph. The graph captures the sequence, the context switches, the information sources consulted, and the implicit reasoning patterns that connect them.

Trailblazer works with three other agent types:

Scout agents run InfraNodus network modeling — gap finding, associative trail mapping, and matchmaking capabilities. They identify structural opportunities in the workflow graph, finding patterns and shortcuts the human doesn't consciously recognize.

3x3 agents provide the analytical framework — three perspectives, three scales, three timeframes — that turns raw workflow data into structured methodology.

MetaTotem provides the persistent memory layer — tracking individual knowledge, preferences, and working patterns across sessions.

The Polytopolis Model

Trailblazer establishes an environmental state space snapshot of all active systems, markets, rules, and processes relevant to whatever you're working on. This snapshot gets visualized using a Polytopolis-inspired model.

You're represented as a series of bubbles. Each bubble shows edges between your current state, sub-optimal states (where things could go wrong), and optimal states (where the highest value outcomes live). Preferences, goals, outcomes, risks, and constraints are all visible.

Trailblazer works to nudge your interactions and deploy your resources toward higher shared outcomes — not by overriding your judgment, but by surfacing the connections and opportunities that your flow state captured but your conscious mind didn't register.

The tour manager doesn't tell the artist what songs to play. It makes sure the right people are in the room, the logistics are handled, and the opportunities that emerge from each performance connect to the next one.

Same architecture. Different domain. Same problem: capturing the intelligence embedded in action before it evaporates.