Build your own team of ten AI assistants, one useful job at a time

Build capacity around a real job

Researcher, planner, builder, reviewer: turn the work you keep putting off into a set of clear jobs you can learn to delegate. The aim is to make a small experiment ready for someone to try. Ten assistants means ten clearly defined roles, not a requirement to buy ten accounts or run ten agents at once. Several roles can begin as separate conversations in one tool. Add automation only after you understand the job and can recognize an acceptable result.

Ten roles to adapt to your work

1. Research scout. Collect sources and questions about one problem you are exploring.
2. Customer-note organizer. Group real feedback while retaining the original evidence.
3. Experiment planner. Write the smallest useful test and a stop or continue rule.
4. Project organizer. Break the test into a few tasks that fit your available time.
5. Writing partner. Turn your brief into clear page copy and instructions.
6. Design partner. Explore a usable layout and call out missing states.
7. Build partner. Help make the scoped prototype and explain unfamiliar code.
8. Testing partner. Try the agreed path and report reproducible problems.
9. Traffic planner. Suggest where relevant people already seek help and draft a useful invitation.
10. Learning librarian. Keep costs, decisions, failures, and reusable lessons together.

Give each role a working agreement

Specify the inputs it may use, the output format, the quality bar, the tools it needs, and where it must stop for your decision. Keep an approved project brief and source folder shared across the workflow. Begin with draft-only or read-only tasks; review a few examples before allowing changes. An assistant needs enough context to help, but not access to every file or account you own.

Practice one handoff before building a team

Start with the first two roles on a real example. Review the output before passing it onward. Ask the second role to flag missing evidence and contradictions rather than quietly filling them in. Keep a short decision record alongside the artifact. If you are correcting the same error repeatedly, improve the brief, examples, or checklist before adding another assistant.

A starting prompt

Use my experiment brief and actual notes. Propose one next test I can complete this week. Include the audience, assumption, smallest artifact, evidence to collect, time and cost limits, and a stop rule. Mark missing facts. Do not invent customer feedback or publish anything.

What would “20×” actually mean?

Define the unit first: for this workflow, it could be a reviewed experiment report. In a hypothetical comparison, 200 minutes of manual work divided by 10 minutes of total AI-assisted work would be 20× faster for that task. The assisted time must include briefing, review, corrections, and an appropriate share of setup. Both results must meet the same acceptance standard. This is an illustration of the calculation, not a result we have measured or a promise about your work.

Measure the whole system

Record at least a few comparable runs before making a performance claim. Track accepted output, total time, tool costs, failures, and rework. Large gains, including more than 20× on a narrow repeatable task, are mathematically possible, but the size and reliability must be demonstrated. Ten simultaneous drafts do not make the business ten times more effective if you spend the next day fixing them. Revenue, customer outcomes, and hours saved are different measures.

Choose tools by the job

• ChatGPT Work: multi-step work with connected context and tools.
• Claude’s Cowork capabilities: help with multi-step knowledge work and files.
• Claude Code: coding tasks with specialized subagents.
• Grok Bot: persistent assistants working across connected tools.
• OpenClaw: an advanced option for separately configured agents and routing.

Access and limits depend on the product and plan. OpenClaw also involves configuration, maintenance, and model costs. You do not need every option. Choose one that fits your current task and learn to review its work before expanding.

Your first-week experiment

Choose one recurring task and capture how you currently do it. Set up the two roles that remove the most repeated effort. Use them on a real piece of work, inspect the result, and record total time and changes needed. Keep the roles that help you produce acceptable work; revise or remove the ones that add overhead. The exciting part is gaining the confidence and capacity to attempt useful work you previously postponed.

Learn the process by building something small

In Two Mondays, Ashley will build a product and show the steps from validation and planning through design, coding, testing, and marketing. Bring a small project and practice the workflow. The workshop meets October 26 and November 2, 2026, from 7–9 p.m. Eastern, on Zoom with a West Palm Beach in-person option. The $99 ticket covers both Mondays and optional setup help and office hours. Paid software and hosting are separate; you can follow the demonstrations without buying every tool mentioned here.

Plan your first experiment

Official tool references

• ChatGPT Work overview
• Claude Cowork overview
• Claude Code subagents
• Grok Bot overview
• OpenClaw multi-agent routing

Tool descriptions checked October 10, 2026. Interfaces, plans, limits, and availability can change. Examples and exercises in this article are original teaching scenarios, not measured client results.