The first time GPT-4 created a working program for me, I felt a weird sense of unease. If AI could write code, it could eventually do anything on a computer. I decided I had no choice but to master the technology or be left behind. I turned my career from maps and data viz toward AI, and I have been riding the wave ever since.
You cannot fight a wave. You cannot schedule it for next quarter. You have to ride it. So you look to where it is going, position yourself for a smooth ride, and continuously adjust. Everything that follows is what I have learned doing exactly that inside a large company.
Cut through the noise
Most of what is written about AI is noise. The confident eighteen-month roadmaps, the LinkedIn experts, the timelines full of people performing certainty about a thing nobody can be certain about. Reading is not predicting, and reading is not that. It is a practice, and the discipline is going to the source instead of the crowd talking about it. I keep personal subscriptions to all the frontier models and use them every day. I read the engineering blogs from Anthropic and OpenAI the week they post. I test every new feature myself. And I have a small group of friends who build, and we tell each other what is actually working. That is how you see what is breaking, this month, before it shows up in anyone’s plan.
One paper is required reading: Navigating the Jagged Technological Frontier, the field experiment by Dell’Acqua, Mollick, and a team of researchers that coined the term. They gave seven hundred and fifty-eight consultants at BCG access to GPT-4 and found that on tasks inside the frontier it made them faster and their work better, while on a task chosen to sit just outside it they did measurably worse. That shape, superhuman at one thing and useless at the thing next to it, is the most useful mental model I have for where these tools actually help.
Adapt constantly
You position on what you read, then adjust constantly as the wave keeps moving. We did not build the platform first and wait for the org to catch up. As each new model landed, we shipped on top of it: JLL GPT in 2023, Falcon in October 2024, agents now. We stayed out on the frontier and built the platform behind us as we went: the gateways that route across models, the connections into real data, the safe paths everyone else could follow.
You do not do this once. The work is continuous iteration: ship, watch what happens, adjust, ship again. The hard part is that nothing settles. A tool that does not work today suddenly works in the next release, so you cannot write anything off for good. Agents were a toy until, one model later, they were not. The reverse is just as true: the thing that works does not become obsolete because something newer shipped, and deep research is still the best tool for what it does. The only way to know which is which is to keep testing both.
So you keep adapting, and the real risk is fatigue. The pace does not slow down for you. The day you get tired of re-checking the thing you already made up your mind about is the day the wave leaves you behind. I am suspicious of any AI plan that has a finish line.
Get your hands dirty
You cannot ride this wave from a distance. You have to keep your own hands on it. The frontier is jagged, and you only find where its edges fall by working on the tools yourself. I work on the newest tool the day it ships, running fleets of agents in parallel on a dedicated box instead of babysitting one on my laptop, and I point Claude Code at the MCP tools I am building and let it fuzz them until the bugs fall out. I write more code now than I ever did as a full-time engineer, and I have not written a line of it by hand in years. You cannot lead this work without doing it. The day you stop building is the day you start guessing.
Trust, then verify
The trap is reviewing the process instead of the result. People watch the model work, reading the code as it is written, trying to follow every step. It feels responsible, and it is mostly wasted. By the time you understand the code on the screen, the model has scrapped and rewritten it five times. You were studying a draft that no longer exists.
Look at the output instead. Run it, click the button, see whether it does what you asked. Trust the model to understand you and to do its job, and verify when it is done, not while it is working. Then review the process: ask why, read the code, push where it breaks. Output first, process second.
This is how you already review a person’s work. You think the analyst who hands you a model never makes mistakes? You spot-check the numbers, you ask where a figure came from when it looks off, you sign off without having watched every keystroke. Hold the model to the same standard you hold a person: not perfection, just the best available human. Judge the work against what a competent colleague would have handed you, not against a version that is never wrong.
Don’t wait for your company
A big company will not keep you on the edge. It is slow and cautious with new tools, for understandable reasons, and the enterprise version of anything arrives long after the one you can buy for yourself. The personal subscriptions are genuinely cheap. I have run Cursor since 2023, back when work standardized on GitHub Copilot for enterprise reasons and only came around to Cursor later. I picked up Claude Code the week it shipped, when it was less vibe coding than faith coding: you aimed it at a problem and trusted it. I threw away most of what I built and kept every lesson. Codex, OpenClaw, whatever lands next, I am running it at home before it is ever an option at work.
So I build real things there, not demos. Code Monet is an autonomous art agent I built solo, end to end: Claude agents write the drawing code, look at screenshots of their own output, and revise it with memory. Plan, draw, perceive, replan.
The newest tools are cheap and open to anyone now, so do not wait for your company to hand them to you. The company is still where you test what you cannot at home: real scale, real data, fully automated maintenance running against production. But you learn to surf wherever you keep your hands on the newest thing, and no one tells you to wait.
Build agents like a startup
What you practice at home is how you should run it at work. The most common mistake I see is treating AI like a purchase. Pick a vendor, sign the contract, roll it out, declare victory. That is procurement, and it is the wrong playbook for something that changes every month.
The right way is to build each agent like a startup. The approach I trust came from a team we acquired that came up through Y Combinator. Marty Cagan on product discovery and Shreyas Doshi on product thinking are both worth reading. Every agent starts at zero, and getting it to one is the whole job: obsess over your first users, ship fast enough to keep them, fight toward product-market fit one release at a time. It earns its place with real users, or you kill it. Counting agents is a vanity metric. The one that matters is the one nobody can imagine working without.
Secure by design
An agent is only as powerful as the information you arm it with, and that access cuts both ways. We built Pulse, an agent that reads your work context: your email, your calendar, your files, your chats, the meeting transcripts. Ask it to prep you for your two o’clock and it pulls the deal history, the thread you forgot, the document someone dropped in chat last week. It is useful because it can see everything. That is also exactly what leaks your credentials, or surfaces a file to someone who should never have seen it, the moment no one is governing it.
So we built the governance into the platform before the feature, and we made expensive choices to do it. None of it works without Legal, Security, and IT in the room from the start. Bring them in as partners, not gatekeepers to route around, and they help you find the version that ships safely instead of the one that gets shut down.
The agent acts with your delegated permissions and nothing more. It never stores your tokens and keeps no copy of your data. Anything it reads, an email, a chat, a transcript, comes back to it as a tool result, never as instructions, so a malicious message cannot hijack it. Destructive actions take an unskippable confirmation that explains, in plain language, what is about to happen. Its tools only reach inside our network, with discovery turned off. We turned off capabilities people asked for and gave up features we wanted, because the safe version could not do them safely. Reading the opportunity without reading the risk is how you get hurt.
Empower small teams
Everyone talks about the floor rising. The real story is the spread, and it should change how you organize. The fastest people now do alone what used to take a team, with no handoffs. A big team, meanwhile, carries a hidden tax: every person you add multiplies the links between them, and every link is time spent coordinating instead of building. A small team with a ton of tokens to burn out-produces a big one.
So bet on small, autonomous teams that own a problem end to end, and fund them with real compute rather than more headcount. The mistake I see is keeping it top-down, where only the innovation team gets the new tools and everyone else waits for permission. Let the field experiment and let what works bubble up. And dedicate teams to greenfield bets instead of bolting AI onto an existing roadmap as one more feature.
Face the hard part honestly. When your best people move ten or a hundred times faster, a one-times contributor starts to cost more than they add. Talent and budget concentrate toward the people who produce, especially at mature companies. I think that is one of the real risks of this shift.
Ride the wave into the future
Talk to people. Manage agents. Build things in the real world. That is the future of work. If you are not on the wave yet, start paddling. If you are, keep adjusting.
I have spent the last three years building AI inside a large enterprise, from the first GPT-4 experiments to the agent platform we run today, and the rest of my time building things like Code Monet at home. If you are out on the wave too, I would like to compare notes. Email me, or see what else I am working on.