We Are Living Through a Technological Revolution

I've been in technology long enough to have a decent sense of what's noise and what's signal. I've seen trends come and go. I've watched hype cycles play out, cool off, and sometimes — rarely — turn out to be genuinely transformative.

Agentic AI is the real thing.

I don't say that lightly. I say it because I'm watching it happen every day — inside my engineering organization, across the teams I work with, in how we think about what software engineering even means anymore.

What "Agentic" Actually Means

There's a lot of loose talk about AI agents. Let me be specific about what I mean, because the specificity matters.

An agentic AI system isn't just a model that answers questions or generates text. It's a system that can plan, act, observe the result, and course-correct — in a loop, autonomously, over extended periods of time, toward a goal you give it.

It can use tools. It can write and execute code. It can call APIs, read documentation, search the web, file bugs, open pull requests. It can coordinate with other agents. It can manage its own memory and context across sessions.

This is categorically different from what most people think of when they think "AI assistant." We've crossed a threshold.

What I'm Seeing on the Ground

At Uber, we're building for massive scale and global complexity. The engineering problems are genuinely hard — distributed systems, marketplace dynamics, real-time matching, multi-modal travel integration across dozens of third-party partners in every timezone.

And increasingly, AI agents are working alongside our engineers on these problems. Not as autocomplete. Not as documentation search. As actual collaborators that can hold context, reason through tradeoffs, generate working implementations, write tests, and flag things that don't look right.

The velocity shift is real. We are moving from idea to global launch faster than we ever have before — not because we hired more people, but because the nature of what a single engineer can accomplish is changing underneath us.

The bottleneck is no longer "can we write this code." The bottleneck is now "can we think clearly about what we want to build."

That's a profound shift. And most organizations haven't internalized it yet.

Why This Is Different From Past Automation

Every generation of engineering tools has promised to make developers more productive. Most delivered incremental improvements. Agentic AI feels different for two reasons.

First, it operates at the reasoning layer, not just the execution layer. Past automation tools helped you do things faster once you knew what to do. Agents help you figure out what to do. They can explore a problem space, propose approaches, identify edge cases you didn't think of, and iterate on solutions without waiting for you to prompt each step.

Second, it compounds. An agent that can write tests makes it safer to move faster. Safer movement means more ambitious bets. More ambitious bets produce more complex systems. More complex systems generate more surface area where agents catch issues earlier. The feedback loops accelerate.

This is why I think of it as a revolution rather than a productivity improvement. Productivity improvements are additive. Revolutions are multiplicative — and they change what kinds of things are even possible.

What This Means for Engineering Leadership

My job is changing. The job of every engineering leader is changing.

For most of my career, engineering leadership has been about three things: prioritizing what to build, building the team that can build it, and removing the friction between them. Those things still matter. But the ratios are shifting.

When agents can draft, test, review, and deploy meaningful chunks of software autonomously, the premium on human judgment goes up, not down. The judgment about what to build, why it matters, what tradeoffs are acceptable, what failure modes to anticipate, what the user actually needs — that's where human leadership becomes more valuable, not less.

The leaders who thrive in this next era will be the ones who learn to work with agents fluidly — who can direct them with precision, evaluate their outputs critically, and combine their speed with human clarity about what actually matters.

And the organizations that thrive will be the ones who restructure around this reality rather than layering AI tools on top of existing processes.

The Honest Uncertainty

I want to be clear about what I don't know.

I don't know exactly where the ceiling is. I don't know how quickly the models improve from here or how soon some of the current limitations get resolved. I don't know which specific roles and functions see the most disruption, or on what timeline.

What I'm confident about is the direction. The capability floor is rising faster than most people expected. The rate of improvement is not slowing. And the compounding effects — better models enabling better tooling enabling better training data enabling better models — show no signs of stopping.

We are not at the beginning of this. We are not at the end. We are somewhere in the steep part of the curve, and the view from here is genuinely remarkable.

---

I've spent my career building technology that moves people through the world. Airports, cities, mountains, continents. Physical movement. Real-world impact.

The technology that's moving through our industry right now is doing something similar — it's compressing the distance between an idea and its realization. Between what we imagine and what we can build.

That's worth paying attention to. That's worth getting right.