Friday, July 31, 2026

{{first_name | there}},

I had planned to use this week's issue to share a detailed map of the agentic systems I use across content creation, project management, and internal operations.

The goal was to show the art of the possible: the relevant skills, how the file system is structured, what goes into and comes out of each system, and how the loops are designed.

That issue needs diagrams, screenshots, and a serious production pass to be useful. Client work this week took priority, so I'm holding off on that one until I can build it properly. This shorter field note looks at a constraint that sits one level above those systems.

The old version of procrastination was obvious. You avoided the proposal and reorganized your desk. You delayed the hard conversation and cleared your inbox. The substitute activity might have felt responsible, but nobody confused it with the real priority.

Agentic tools have changed the camouflage.

Now you can avoid the hard decision by launching a research sprint. You can turn a loose thought into a polished memo, build a working prototype, compare twelve tools, or produce a complete project plan before deciding whether the project matters. The output can be real, useful, and impressively finished. It can also be the wrong work.

Peter Yang recently described three habits he's watching in his own use of AI: relying on summaries instead of reading the originals, letting agents create new forms of distraction, and sometimes preferring the agent interaction to dealing with another person.

I think his framing is important because it names the human and social costs directly. My contribution is adjacent. I'm interested in the operating cost that appears when delegation becomes nearly frictionless.

Starting used to impose a tax. You had to open the blank document, gather the material, or convince another person that the work was worth doing. That friction killed plenty of good ideas, but it also killed weak ones before they consumed a week.

Agents remove much of that tax. A half-formed impulse can become a substantial body of work before you've made a real priority decision.

That is what makes productive procrastination hard to catch. The artifact looks like progress. The agent followed the instructions. The work may even become valuable later. But none of that matters if what it displaced was more important or urgent.

I was outlining this issue with the help of a voice AI agent, using it as a thinking medium: a way to explore directions, generate ideas, and hear the shape of an argument before committing to it.

That's what makes voice interaction with agents such a fun new medium. It can feel like having an always-available brainstorming committee. You can throw out a partial thought, ask for another angle, follow an association, and test three directions without asking anyone to find an hour on their calendar. It's fast, useful, and genuinely enjoyable.

The tradeoff is less obvious. Conversation lets you begin before you've structured the problem. That's often the point, but it can also reduce the silent, solitary work of deciding what you're trying to solve, defining the objective, and choosing what actually needs to be communicated.

For some work, that deeper structuring is the work. It matters for high-stakes decisions, important presentations and decks, client deliverables, and handoffs where another person needs to act without the whole conversation in their head. A written memo forces the objective, decision, scope, and message into a form someone else can inspect. And taking the time to write it without the help of an eager brainstorming committee can be a useful forcing function for deeper thinking.

I use voice and conversation for exploration and brainstorming. I use written memos that I write myself for commitments, decisions, scope, and handoffs. Neither mode is universally better; they have different jobs.

The risk appears when the exploratory mode quietly authorizes execution. With an agent in the loop, every new direction can become a research task, a document, a deck outline, or a build. A productive conversation can generate polished evidence of progress before you've fully settled on what you're driving towards.

Sometimes the conversation leads to a dead end or just lays out the options worth considering. In those cases, it has done its job: it helped you understand the idea well enough to stop. The problem starts when every branch becomes another task to keep moving.

Agents let you work on many things at once. Start a research task, move to a deck, send another agent after the next idea. Each thread progresses while your attention jumps.

The cognitive load is easy to miss because execution has left your screen, but the hard decisions haven't. You still own the objective, assumptions, and quality bar. As the number of concurrent tasks grows, those decisions pile up on you.

Eventually, bandwidth becomes the constraint. You have less time to pressure-test each output, catch weak assumptions, or bring the same quality to every thread.

Good systems absorb some of that load. With the right context, decision rules, and feedback loops, an agent can make more choices without routing every question back to you. Thin scaffolding does the opposite: garbage in, garbage out happens in parallel.

Multitasking becomes procrastination when the number of active threads outgrows your ability to make the hard decisions well. You put yourself in a position where you can't bring your best judgment to the work.

Agents can keep going. Your quality bar depends on whether you've built enough scaffolding and left yourself enough bandwidth for the decisions that remain.

— Jamaur