Agents, plainly · Part 3 of 3 · Aug 2026
What professionals actually do with agents now
Every week I talk to people who run agents and people who assume ChatGPT is the whole story. The distance between them is already large. Here's how the first group actually works.
Knowledge work, handed off
Claude Cowork and ChatGPT Work exist for one move: hand over a messy goal — “brief me for Tuesday's call from these notes” — and come back to finished work for review. Drafts, digests, meeting prep, first-pass research. The agent works in your files and returns a deliverable, not a paragraph.
Inbox, scheduling, and follow-ups
This is my world. An agent with memory and mail access triages, drafts, schedules, and chases follow-ups; the human reviews. Loopsmate does the same job for the open loops running in your head. It is not ChatGPT asked to write a nicer email; it's a system that carries the state of your obligations.
Repetitive ops and internal tools
Professionals who code — or got close enough — run terminal harnesses (Claude Code, Pi, dsh) on their own automations: the Friday report generated by itself, the data pulled and sanity-checked, the script written and run. The pattern to notice: agents write the tools now.
Always-on teams
Grok Bot is the loudest sign of the next pattern: agents that hold their own computer, operate apps, and keep working 24/7. Early beta, priced for businesses, weeks old as I write this. Skepticism is warranted; the direction is not.
The gap, from inside
Thomson Reuters' survey of professionals found AI embedded in everyday work but a widening gap between what firms expect and what they can execute. I see the same thing one level down: the people handing multi-step jobs to agents are compounding — hours back, briefs written, loops closed — while everyone else is still asking chat a question. The agent era is the first one where the tool itself widens the gap. Usable is the work.
That's what I build: the harness, the partner in Telegram, and a managed agent service for professionals. If you want your week back, we should talk.
Sources: Anthropic · OpenAI · xAI · VentureBeat · Thomson Reuters