The arc, through one question
"We get 200 customer emails a week. Can AI help?" Here's how each phase changed the answer.
- Foundations (Days 1–14): a model learns patterns from examples. Measure it with precision and recall, and test on data it never saw.
- Large language models (Days 15–28): tokens, context windows, prompts and JSON output. Now it can read an email and summarize it, and you know it can hallucinate.
- Knowledge and RAG (Days 29–34): ground answers in your own policies with retrieval, so replies cite real sources.
- Agents (Days 35–46): tools, memory, planning, MCP, human approvals and guardrails against prompt injection. Now it can act, carefully.
- Builder toolkit (Days 47–54): terminal, git, Python, APIs, keys kept secret, your first LLM call.
- The tools (Days 55–84): GitHub Copilot, OpenAI Codex, Claude and Microsoft Copilot Studio. Different strengths, same ideas underneath.
- Capstone (Days 85–96): MeetingMate, from transcript to action items, with tools, knowledge, evals, cost checks, approvals, a deployment and a README.
- Bigger picture (Days 97–99): real patterns at work, responsible AI, and a 90-day roadmap.
From "can AI help?" to "here's what I'd build, how I'd test it, and where the human signs off."
Three lessons I'd keep if I forgot the rest
1. Start with the problem, not the tool.
2. If you can't measure it, you can't trust it. Write tests first.
3. Keep a human in the loop where money, health or people's rights are involved.
What's next
The tools will keep changing. The ideas won't: models predict, retrieval grounds, tools act, people decide. You can now read each new release and know where it fits.
Your action today
1. Pick the one day that helped you most and share it with one colleague.
2. Write your 90-day problem from Day 99 in the comments, so we can hold each other to it.
3. Come back to any day when you need it. All 100 posts stay up.
Where to go from here: your own work. Banking, insurance, healthcare, retail, product, analysis. That's where the next 100 days of learning happen.
In one sentence
You started with "what is AI?" and finished able to design, build, test and ship an agent responsibly.
Thank you for every comment, question and correction along the way. They made this series better. Keep building, and keep sharing what you build.
What was your biggest "aha" moment in these 100 days? 👇