The Signal

The Lead

The encoding is the work

I've written three posts over the last few weeks, each prompted by something different. The first was about project managers adopting AI before writing down what they actually decide. The second was about agent authority — the question of who an agent is acting for, and on what basis, the moment Microsoft and OpenAI started shipping autonomous workflow agents. The third, this week, was about digital twins — specifically the trend of LLM personas being demoed as twins and recommended to audiences for serious decision work.

Three different prompts. Three different surfaces. Same underlying argument every time.

Across all three, the same shape kept appearing. There's a visible layer — the tool, the agent, the persona, the dashboard, the orchestration platform. That's the layer that gets demoed in a webinar, written about in an analyst note, and bought against an annual budget. It's the part anyone can see. And there's an invisible layer underneath it — the encoded judgment, decision criteria, exception logic, escalation rules, and 3am-on-the-shop-floor knowledge that has never been written down. That's the layer that determines whether the visible one does anything useful.

The visible layer is moving fast. Stanford's 2026 AI Index reported this month that agents jumped from twelve percent to sixty-six percent success on real computer tasks in a single year. Google launched the Gemini Enterprise Agent Platform at Cloud Next. Adobe rebranded Experience Cloud as CX Enterprise and started selling "Coworkers" that orchestrate workflows toward business goals. The capability is here. The capability is going to keep getting better.

The invisible layer is not moving. It moves only when somebody sits down with the people closest to the work and writes down how decisions actually get made. Not how the org chart says they get made. How they actually get made. The criteria. The exceptions. The judgment calls. The things that are obvious to a senior operator and opaque to everyone else. That work is slow, human, political, and impossible to demo. And it's what determines whether any of the visible-layer capability turns into outcomes.

Three weeks of writing made this concrete in a way reading about it didn't. The PM post argued that tools amplify the logic you already have, and produce noise when the logic is implicit. The agent authority post argued that workflows can't be delegated to agents until someone has written down what those agents are allowed to decide. The digital twin post argued that a twin without encoded behavior isn't a twin — it's a costume on an LLM. Different vocabulary in each piece. Same gap underneath.

We call this work Logic Extraction at ResonAi, and it's Phase Two of the methodology — between Discovery and any tooling decision. It is the unglamorous middle of every AI engagement that actually compounds. It's also the part most engagements skip, because it's the part nobody can put on a slide.

The encoding is the work. The work is the gap. The gap is the moat. Every implementation that's actually shipping value this year is on the right side of that line. Most of the spend isn't.

From the Practice

The three-different-heads problem

Every Logic Extraction engagement runs into the same surprise about a week in. The decision the team thinks they're encoding lives in three different heads, and those three heads have never reconciled the rules they're each using. Until the work starts, everyone assumed there was a single answer. There isn't. There are three answers, weighted differently by whoever happens to be on the call.

This is also why Logic Extraction is political work disguised as documentation. Encoding the decision means picking which head's version becomes the rule, and that's a real organizational move. Skip the picking and the agent ends up running the version no one would have approved on the record.

If a Logic Extraction engagement isn't surfacing the three-different-heads problem within ten days, it isn't being done close enough to the work.

If your team is heading into an AI rollout — agent platform, copilot, vertical tool, twin, anything — and the answer to "where does our decision logic live" is somewhere between "in our heads" and "we'll figure it out as we go," that's a Discovery call worth having before procurement, not after.

Or read this issue on LinkedIn.