June 2, 2026
A developer named Marcos started with a completely blank repository. No configuration files, no automation, nothing.
He typed one command: /semai init
From there, an AI agent read the codebase, figured out the tech stack, and generated a full CI/CD pipeline tailored to what it found. It recommended linting, security scanning, matrix testing, topology improvements.
Then Marcos told it: "Work until the pipeline is green."
The agent monitored the pipeline, analyzed failures, applied fixes, re-ran workflows, and kept going until the build passed. Then it summarized everything it had changed so the developer could actually understand what happened.
If you want to watch the whole demo, it's here.
But here's what caught my attention. And it wasn't the pipeline.
What's Actually Happening
There are two things working together in that demo that make it more than a clever trick.
The first is the slash command. When you talk to an AI in freeform English, the agent has to interpret what you mean every single time. That interpretation is inconsistent. Sometimes it does what you expect. Sometimes it doesn't. You can't predict which, and you can't repeat it reliably.
A slash command like /semai init is different. It's a named, explicit entry point that maps to a defined workflow. Think of it as the difference between a verbal briefing and a runbook. The runbook wins every time, at scale.
The second ingredient is context. The agent didn't just run a generic script, it read the environment first and understood what was really there. A consultant who spends a week reading your files gives you better advice than one who shows up cold. Same principle.
The combination is what makes "work until the pipeline is green" possible. Reliable entry point. Rich context. A collaborator that stays in the problem until it's solved, then hands you a clear summary of what changed and why.
That last part matters more than people are giving it credit for. The summary isn't a bonus feature, it's how the human stays in the loop. It's how knowledge transfers. Without it, you have a black box that did something. With it, you have a process you can learn from, repeat, and hand off.
This Isn't a Developer Story
Most of the people I work with will never touch a CI/CD pipeline. That's not their world.
But they know every single problem this demo is solving.
Processes that break quietly and nobody notices until it's too late. Onboarding that falls apart because someone skipped a step and there was no one to catch it. Staff who can't find the answer they need, give up, and either do the wrong thing or ask someone who's already stretched thin.
The pattern in that demo is a people pattern. A "how do we make sure things actually get done, correctly, without requiring heroics" pattern.
The tech is just where it showed up first, because developers tend to build the things they need.
What It Looks Like for a Nonprofit
Let me give you some concrete versions of this.
A new staff member joins your organization. They're walked through onboarding by a helper that knows exactly where they are in the process. Not a static checklist they print and lose. Not an email drip that fires on a schedule regardless of what they've done. A system that knows step 4 wasn't completed and sends a specific, contextual reminder: "You haven't uploaded your direct deposit form yet. Here's the link."
I have three appointments for my daughters this week. I have so far received four specific reminders, each tied to exactly what I still needed to fill in. Without those reminders, I would have missed all of it. I simply wouldn't have thought, "oh let me log into three different portals and check what documentation I haven't completed yet", or "oh let me reference the three emails I received a week ago with the multi-steps of paperwork listed across four paragraphs of instruction in each". The reminders aren't a luxury feature. That's just how things get done now, and the organizations that build their processes around it are the ones whose staff and constituents actually complete what they start.
Or take something simpler. A staff member needs to run a report. They don't know where it is, they don't remember what it's called. Today that's a ticket, a search through documentation that may or may not be current, or a message to whoever has bandwidth. Tomorrow it's a question in plain English: "How do I pull this month's program data?" And the answer comes back in seconds, pointing them exactly where to go, or just doing it.
These aren't hypothetical futures. All of this is happening, in pieces, right now. The organizations that are building toward it intentionally are going to look very different in three years from the ones who aren't.
What We're Watching at Coat Rack
I had a conversation recently with a VP at one of our clients. He wanted to know how we're using AI, where it's added time, where it's created problems. It was a good conversation, and it made clear that nonprofit leaders are paying attention. They're not waiting for a whitepaper. They want to know what's actually working.
We are watching closely how AI changes the experience of working in and with a system, not just the experience of building one.
The onboarding and support gap is real for a lot of the organizations we work with. They've built systems that work, but the experience of learning those systems, navigating them, and getting unstuck hasn't kept up with what people now expect. That gap is closeable. The technology exists. What it takes is intentional planning and someone to help you think through where to start.
That's exactly the kind of conversation I like having.
One Next Step
If you read this and thought "we have that exact problem," let's talk. An initial conversation is usually enough to figure out whether there's a clear path forward or whether it's too early to act.
No pitch, no deck. Just a conversation. Schedule a free consult here.
Lisa Montague is the CEO and co-founder of Coat Rack, a nonprofit technology consulting firm.
