There’s something I’ve been wrestling with lately. A thought that keeps popping up in meetings, airport lounges, and those late-night work sessions where you suddenly realise you’re still staring at the same notebook page.
It’s this:
Public transit doesn’t actually need more technology. It needs more hands. Not literal hands. Not more staff. It needs more capacity, the operational kind.
And AI agents are quietly becoming the closest thing we’ve ever had to adding 20 extra operations specialists without hiring a single person.
Let me break down why this matters, what transit agencies, transit suppliers, or founders should be building, and why the next wave of transit innovation won’t come from dashboards… but from autonomous “digital operators” sitting inside the system.
We Don’t Need More Dashboards. We Need More Doers.
Let me be straight with you.
I’ve never met a transit leader who said, “We just need one more dashboard and we’ll magically hit our KPIs.”
Never. Not once.
The real problem is simpler and more painful:
There’s too much work and not enough people to do it.
Transit runs on:
constant coordination
endless micro-decisions
reactive firefighting
late-night rescheduling
manual data chasing
incident triage
energy juggling
maintenance juggling
“Who’s available?” juggling
It’s juggling every direction.
And humans can’t operate at that scale anymore. Not with ridership swings. Not with electric fleets. Not with urban complexity rising every quarter.
That’s why AI agents matter. Not because they’re “cool”, but because they take work off people’s plates.
For the first time in public transit’s history, we aren’t just getting insights. We’re getting helpers.
AI Agents Aren’t Tools. They’re Teammates.
Here’s where I think people misunderstand AI agents.
They still think of them as software features. Or plugins. Or something the IT team deploys after a long procurement cycle.
That’s not the right way to see this.
AI agents behave like junior ops staff. Fast. Consistent. Tireless. Sometimes a little too literal. Sometimes needing supervision. But they show up, do the work, and don’t complain.
Agents don’t “report insights.” They take action. They update schedules. They resolve small incidents. They re-route services. They communicate with passengers. They distribute charging loads for electric fleets. They log operational decisions in real time. They escalate only when something smells wrong.
This isn’t theoretical. Cities testing early versions have already reported measurable improvements in reliability and response time (UITP 2023; McKinsey 2024).
AI agents are the missing operational layer we’ve been pretending humans could handle on their own.
They can’t. Not anymore.
The Night I Realised AI Agents Had Crossed a Line
A few months ago, I was talking to the operations director for a mid-sized city with a surprisingly complex bus network. Think 1,200 vehicles, unpredictable weather, and more construction zones than anyone wants to deal with.
We were testing a simple scheduling agent. Nothing fancy.
One night, around 11:30 p.m., the agent flagged three potential issues for the following morning:
A driver shortage on a specific corridor
A cluster of predicted delays due to expected rainfall
Two buses are showing early signs of battery degradation
Normally, this would’ve waited until 6 a.m. And then exploded into stressful morning chaos.
But here’s what the agent did on its own:
It created a new schedule. Reassigned vehicles. Updated driver rosters. Adjusted the deadhead routing. Allocated backup vehicles. Notified the depot supervisor. Triggered a maintenance ticket. Updated the passenger ETA model. Pushed a preemptive “expect minor delays” alert.
All before midnight.
I sat there staring at the logs and thought, “Okay… something just changed. For real.”
It wasn’t perfect. It wasn’t sexy. But it was operationally meaningful.
And that’s what matters.
Framework: The AI Agent Stack (Simple, Not Over-Engineered)
When founders ask me, “What agents should we build for transit?”, I give them a framework that’s not fancy, but it’s brutally practical.
Think of the stack like an operations team.
1. Frontline Agents
These handle fast, light, repetitive decisions. • delay classification • rerouting • schedule tweaks • passenger notifications
Small tasks. Big impact.
2. Specialist Agents
They own one domain deeply. • charging optimisation • predictive maintenance • asset health monitoring • real-time fleet balancing • energy load orchestration
Cities are already piloting these with promising results (TfL 2024).
3. Coordinator Agents
These connect everything together. • “If this happens, do that.” • multi-agent negotiations • network-wide reasoning • inter-department coordination • risk scoring and escalation
This is where things start feeling like real teamwork.
4. Supervision Layer
Humans. Always humans.
Agents make decisions fast. Humans control direction and override bad calls.
You never remove humans. You remove the overload placed on them.
Execution: How Founders and Leaders Can Start Today
Alright, now the real stuff. The part people skip because it’s not glamorous.
1. Pick one workflow that annoys everyone
Not the biggest. The most painful. Usually it’s: incidents, timetables, charging, or maintenance.
2. Build one agent that handles the full workflow
Not a model. Not a script. A real agent that starts and completes the flow. Start-to-finish ownership builds trust.
3. Give the agent permission… then boundaries
Let it act. But set the rails. Agents don’t need unlimited freedom. They need clarity.
4. Track decisions in plain language
Agents should talk like teammates. “I rerouted Line 15 because Route B got blocked. Readable. Reviewable. Human.
5. Graduate the agent from “assistant” to “operator”
Make it responsible for outcomes, not tasks. Once teams start relying on it, everything shifts.
Takeaways: The Future Ops Team Will Be Hybrid… and Smaller
Here’s the honest truth I’ve learned watching agencies adopt early AI agents:
• We don’t need more people staring at screens. We need fewer people making smarter decisions.
• AI doesn’t replace staff. It replaces noise, busywork, and manual grind.
• The agencies that win will be the ones that scale operations without scaling headcount.
• The startups that win will build the agents that quietly save millions every year.
And I’ll say this again because I want transit agencies, transit suppliers, and founders to really hear it:
Dashboards won’t fix transit. AI agents will.
A question for you
If you could hire one AI agent tomorrow, like an actual teammate, which workflow would you want it to own first?
Think about it for a moment. Where’s the real pain?
I’d love to hear what you see.
References
Carnegie Mellon University 2024, Surtrac Smart Traffic Research Project.
International Association of Public Transport 2023, Artificial Intelligence in Public Transport.
McKinsey and Company, 2024, AI in Infrastructure and Urban Mobility.
Smart Cities Council 2024, Connected Transit Systems and AI Impact Study.
Transport for London 2024, Predictive Maintenance Case Study.
Top comments (0)