Workflow 1
Build and publish a website with agents
Turn a clear visitor task into a tested page, with a reviewable path from brief to live URL.
Follow the workflow →Ashley Kays · Put AI to Work · From ideas to results
How I Build with AI
The tools, briefs and review steps behind useful work—with three workflows you can adapt.
Workflow 2
Create a useful course with an agent team
Start with what learners should be able to do, then build examples, practice and feedback around that outcome.
Follow the workflow →Workflow 3
Research and test an income idea with agents
Use your skills and customer access to test a small offer before committing to a bigger build.
Follow the workflow →My stack, with a job for each tool
This series builds on the tools and examples in Steal My Stack. The workflows below are reusable teaching patterns. A tool in the wider stack is not a claim that it ran on every project, or that every role was a separate autonomous agent.
| The job | Tools from the published stack | How to choose |
|---|---|---|
| Brief, research and drafts | ChatGPT with GPT-6 Astra; Claude | Use the assistant you can review reliably. Give it the same source brief and acceptance criteria. |
| Code and implementation | Codex; GitHub | Use a coding workflow for source changes and a reviewable version history. |
| Design and prototypes | Figma | Use a visual prototype when it resolves a question about the experience. |
| Publish | Railway or Vercel | Follow the existing project’s deployment path. This site uses Railway. |
| Optional accounts and data | Supabase where needed | Add persistent data only when the product requires it; this is not the database choice for every site. |
| Payments and appointments | Stripe; Cal.com | Use only configured, tested flows for agreed offers. |
| Email and measurement | MailerLite; GA4; Search Console | Keep permission, configuration and actual customer outcomes visible. |
A small team, directed by a person
Start with a lead assistant and a separate review pass. Add specialist workers when the tasks are sufficiently distinct. The human owns the objective, permissions and acceptance decision. The lead keeps one brief, assigns bounded tasks and collects evidence. Each worker returns an output, checks performed, unresolved questions and the next owner.
These pages teach role-based workflows; they do not launch agents or connect to your accounts. You can run the roles sequentially in one assistant or delegate them in a tool that supports it.
The handoff that keeps work moving
- Input: the approved brief and relevant sources.
- Output: a named file, draft or working version.
- Acceptance: what must be true for the next person to use it.
- Evidence: sources, checks and any unresolved defects.
- Next step: who acts next and what approval is needed.
What still needs human judgment
Choosing the problem, speaking with customers, making tradeoffs and accepting the finished work remain human responsibilities. Agents prepare drafts and implementation; they do not make an unsupported claim true. Review the exact material before sending, spending, publishing or making a commitment.
Track the cost of an accepted result
Keep subscription costs separate from variable usage, hosting, creative assets and human time. Allocate shared costs consistently across projects. Record setup, failed attempts and review—not just the successful run. Enter current prices from your own account; this series does not quote subscription prices or promise a productivity multiplier.
Download the three workflow templates
Choose your first output
Make one page, one complete lesson or one experiment card. Once you can review that output reliably, expand the workflow. For guided help, explore working with Ashley or browse training.