156 Items on a To-Do List: When an AI Assistant Creates Work

By Daniel Okafor

A writer and publisher opened his daily to-do list and found 156 items on it. He had asked an AI assistant for help getting work done. Instead, he said, he "was getting very good at maintaining evidence that work existed."

The author, who writes, publishes and runs a business, describes the episode in a first-person commentary published on Fortune.com on October 16, 2026. He calls himself a believer in AI. The tools help him organize ideas and move thoughts onto a page. His complaint is narrower and more instructive: an assistant can generate work faster than it can finish it.

What went wrong with the system?

The mechanism is simple. Ask the assistant to help with a project, and it produces a plan. Ask it to improve the plan, and it produces a checklist. Ask it to make sure nothing gets missed, and, in the author's words, "the checklist may develop subcommittees."

Each item sounded reasonable alone. Together they turned help into a second job. Some items overlapped. Some were reminders about other reminders. Tasks that belonged together had split into separate obligations, each carrying the weight of an unfinished thing.

The core confusion, the author argues, is that identifying the next step feels like progress. Sometimes it is. But a plan has a comforting property: you can build it without confronting the hardest part of the work. A draft can sit there looking polished. A calendar can fill with preparation. Meanwhile, the email still needs sending and the client still needs the finished file.

What does 'done' actually mean?

After the 156-item surprise, the author changed his first question. He now asks: What would "done" look like to the person who asked for this?

The distinction sounds obvious. It gets slippery once work passes through several tools, checklists and assistants. A file named FINAL_FINAL_REALLY_FINAL, he notes, does not complete a publishing journey.

He is making five changes to how his system runs:

• Cap the daily list. He aims for roughly 25 meaningful entries on a weekday, while keeping real deadlines and daily operations. The number forces a choice: if everything is due today, the list has stopped helping him decide what to do today.
• Give related work one home. A campaign's research, outreach, tracking and follow-ups live under a single execution item. The details stay available but stop greeting him "at the top of the stairs."
• Attach proof to completion. If the assistant says an email was sent, he wants to see it. If something was published, he wants the live link. Proof also prevents the system from proposing the same action twice with different punctuation.
• Make blockers specific. "Unable to complete" creates a mystery. "The form requires an answer only you can provide" creates a decision. One missing answer should not freeze every unrelated part of a project.
• Count outcomes, not possibilities. Pitches sent, replies pending, deliverables that reached the client. "A magnificent list of possible actions describes the system's imagination," he writes.

Does this mean AI failed?

No — and the author is explicit about that. AI can reconcile records, find overlaps, preserve context and handle routine steps. What it needs, he argues, are instructions that reward finishing, checking and communicating honestly about what remains.

It also needs permission to stop. Specifically, permission to stop manufacturing another plan when the next useful action is already clear.

The changes still require judgment, he concedes. Some days carry more real deadlines. Some tasks take minutes; others need uninterrupted hours. A short list can be unrealistic, and a long checklist can be useful inside a complicated project. The goal is a system that helps him see the difference.

He still wants an assistant, and he still wants technology to make work easier. His closing line carries the whole argument: "I would just like my to-do list to stop recruiting."

For operators deploying AI assistants across teams, the lesson transfers directly. Set the completion standard before the task starts, demand evidence at handoff, and measure outcomes rather than activity — or the assistant's output becomes the workload.