Your First Session
A narrated walkthrough — from first connection to coached, context-aware building in about 10 minutes.
You've connected your agent. Now let's see what happens when you start talking to it.
What to Expect
Your first session is about establishing your coaching foundation. By the end, your agent will know your goals, your tech stack, your conventions, and the decisions you've made and why — and it will use all of that to coach you, review your work, and validate outcomes in every future session.
Your Agent Discovers the Project
Tell your agent to get started. It reads your codebase — tech stack, file structure, tests, CI — and reports what it found.
# You say:
"Get my status"
Your agent now knows your project. If it's a greenfield repo, it notes that too — you can start from scratch.
You Set Project Goals
Your agent asks what you're trying to achieve. These goals shape how it coaches you going forward — what to prioritize, what to question, what to push back on.
# The agent asks about your goals:
"Increase user engagement and reduce onboarding friction"
Teach It What Matters
This is the most valuable part. Tell your agent the things it should know for every session — the things you're tired of repeating.
# You might say:
"We use RSpec with factory_bot, no fixtures."
"We chose Stripe over Paddle because we need marketplace payouts."
"Rails 8 API backend, React frontend, PostgreSQL with jsonb for flexible fields."
"We tried Redis for caching but it was overkill. Switched to Solid Cache."
Your agent stores each one. From now on, this knowledge surfaces automatically — in every future session, without you repeating it.
What's worth teaching?
- Tech stack decisions and why you made them
- Coding conventions your team follows
- Architecture decisions — and what you rejected, and why
- Domain knowledge the AI wouldn't know
- Past approaches that didn't work (so the agent doesn't retry them)
Your First Coaching Moment
Now start working — a feature, a bug fix, whatever you need. Your agent brings everything it knows to the conversation. But here's where it gets interesting: when you're about to invest real effort, your agent coaches you to think about outcomes.
# You say:
"Let's add a quick-start wizard for new users."
# Your agent coaches you:
"Good idea. Before we build it — what would tell you it worked?
I'd frame this as a bet: we believe a quick-start wizard will get
new users to their first success faster. Signal: time-to-first-action
drops below 5 minutes. Check in 2 weeks. Want me to set that up?"
This is the moment DevKeel clicks. Your agent isn't just writing code — it's helping you define what success looks like before you build. You say yes, and now your work has a target.
As you build, your agent reviews its own work automatically — checking code quality, security, tests, and whether the implementation matches what you intended. You don't have to remember any of it.
The Loop Closes
Two weeks later, you open a new session. Your agent doesn't wait for you to ask — it knows the bet's timeframe is up.
# Your agent opens the session:
"Your quick-start wizard bet is due for a check. Time-to-first-action
was 12 minutes before — want to see where it landed?"
Whether the signal was met or not, you capture what you learned. That learning carries into every future session — shaping what you build next, not just what you built before.
This is the full cycle: bet → build → check → learn. Most tools stop at "build." DevKeel closes the loop.
The Real Payoff
The value isn't in the first session. It's in every session after.
Without DevKeel
"We use RSpec, not Minitest. No, we chose Stripe, not Paddle. The API is Rails, not Node. We tried that approach already, it didn't work because..."
With DevKeel
"Add the subscription webhook endpoint."
Your agent already knows it's Rails, Stripe, RSpec — and that you tried a different approach last month that didn't scale.
Your agent remembers everything — and uses it to coach you toward the right things to build and tell you whether they worked.
Next Steps
You've seen the full cycle in miniature. Now dive deeper into how each piece works: