How agents work
Z2X Founder runs AI agents to do the research, writing, and validation work that normally takes a founder weeks of manual effort. Understanding how agents work helps you get the most out of them.
The approval model
Every agent in Z2X operates within a human-in-the-loop approval model:
- You or Autopilot triggers an agent run
- The agent executes — researching, writing, or analysing
- The agent produces an artifact — a document, report, or action
- You review the artifact and decide to Approve, Request revision, or Reject
- Only after your approval does the artifact become "official" — added to your evidence ledger, used by subsequent agents, or (for outreach drafts) visible as a draft to send
No agent ever sends an email, publishes content, or makes a decision without your explicit approval. This is a core design principle.
Agent types
| Agent | What it does | Credits |
|---|---|---|
| Step Runner | General research, analysis, and drafting | 2 |
| Competitor Agent | Maps the competitive landscape for your idea | 3 |
| Pain Signal Agent | Finds evidence that real people have your target problem | 3 |
| First 100 Source Map | Identifies the exact channels where your first customers hang out | 4 |
| Growth Hacker Agent | Drafts outreach sequences and early acquisition strategies | 4 |
| Builder Agent | Scopes MVPs, generates specs, drafts landing page copy | 5 |
| Decision Gate | Synthesises evidence and recommends continue/pivot/stop | 5 |
Each agent is specialised. The Competitor Agent knows how to structure a competitive analysis. The Pain Signal Agent knows how to identify and weight different types of evidence. You don't need to instruct them — Z2X routes each step to the right agent automatically.
Credits
Running an agent uses AI credits. Credits are consumed when you click Run agent — not when you review the output. See AI credits explained for full details.
What agents can and can't do
Can do:
- Research using publicly available information
- Draft content based on your brief and profile
- Analyse and summarise information
- Score and rank options
- Generate structured plans and specs
Can't do:
- Send emails or messages on your behalf (they draft; you send)
- Access private accounts or systems
- Make purchases or commitments
- Execute code or build software directly
Agents are your research and writing staff. You remain the decision-maker and the person who acts in the world.