如今想真正做好 AI，你得成为智能体部署专家

We’ve just gone through 3 phases with AI that most people don’t fully appreciate:

• Phase one: most of 2023.  You had to be technical. The models were there but they hallucinated constantly. You needed to be deeply technical to get anything useful out of a raw LLM API. Most of us — myself included — weren’t equipped. I remember being at SaaStr Annual 2023, talking with David Sacks, asking how he was thinking about AI at Craft. He said they wanted 80% of investments to be AI. I asked to see the great ones already in market. His answer: they’re all proof of concepts. We’re all in anyway. That was the right call if you were investing at the LLM layer. I wasn’t smart enough to play there, let alone deploy AI B2B agents then.
• Phase two: 2024 into early 2025: the weird prompt engineer era. You could torture these tools into doing something useful, but you had to craft these elaborate, convoluted prompts that made no sense to anyone else. “Prompt engineer” became the hottest job on the planet for about a year. That job is now dead.
• Phase three — which is right now — is the era where ordinarily smart generalists can make AI agents and AI tools do genuinely magical and useful things. No contorted prompts. No engineering degree. Just software deployment skills you probably already have. Some of it is the profound leap forward of Opus 4.5+.  Some of it is the agentic products themselves just have gotten better.  It’s both.  It’s now.

That transition matters enormously.

Because the questions everyone are asking — who do I hire, who do I keep, what does great look like on my team — have a clear answer now.

And it’s not what most people think.

The One Interview Question That Tells You Everything

Stop asking candidates if they’re “familiar with AI tools.” Stop asking what they do with Claude or ChatGPT. Those questions were appropriate in late 2024. They’re not the bar anymore.

The question in 2026 is: what commercial AI agent or tool with real ROI have you brought into your organization in the last 30 days?

Not played with. Not demoed. Not watched a YouTube video about. What did you actually buy, configure, deploy, and put in front of your team in the last 30 days?

Anyone genuinely operating at the cutting edge will light up answering this. They’ll tell you exactly what they bought, why they picked it over the alternatives, how they trained it, what broke in week one, what it’s doing now. They can’t stop talking about it.

The people who aren’t there will stare at you. You’ll see it immediately.

We do a lot of reviews with fast-growing companies — teams crossing $100M and well above — and I ask this question of their management teams regularly. At even the best startups I’ve invested in, maybe 30% of the management team clears this bar. In general interviews, it’s single digits.

That gap is the opportunity and the risk, depending on which side of it you’re on.

The Job Title We Need to Start Using

Prompt engineer is gone. Go-to-market engineer — a title that was everywhere six months ago — is already fading. What we actually need, at every level from C-suite to junior, is what I’d call an agentic deployment expert.

Not someone who builds AI. Someone who deploys it.

The distinction matters. The value in 2026 is not in engineering the underlying models. Anthropic and OpenAI and others have done that work, and they’ve gotten extraordinarily good at it. The value is in identifying which of those products will improve your specific workflows, getting them into your organization, training them properly, and measuring the output.

That’s a software deployment skill. And the people who are excellent at it — who can come in and immediately tell you the eight ways a given tool would improve your team’s productivity and the three ways it wouldn’t — are worth more than almost anyone else you can hire right now.

You Do Not Have to Be Technical

This is the part I want to be especially clear about because I think a lot of smart, capable operators have talked themselves out of the game.

If you felt left behind in early AI — if the conversations about APIs and model weights and fine-tuning felt like a different language — that era is over. You don’t need to speak that language anymore.

Here’s the test: have you, personally, successfully deployed a piece of enterprise software in the last three to five years?

Not hired an agency to do it. Not delegated it completely. Have you personally gotten in, learned the tool, set it up, trained your team on it?

Salesforce. HubSpot. Outreach. Gong. Whatever it is. If the answer is yes, you can deploy any AI agent on the market today. The only genuinely non-intuitive part is training — which is a real step that many teams skip and then wonder why the agent isn’t working. But the UI, the UX, the configuration logic — it’s just software. It works like software has always worked.

The people who are winning with AI right now in non-technical roles aren’t doing anything magical. They tried the tools early. They trained them for a month. They stayed a year ahead of everyone else through one simple mechanism: they actually used the things.

What Happens If You Don’t

I was reviewing a company recently — a well-run team, growing fast, crossing toward $100M. Their new CRO hadn’t brought in a single AI tool. Not one. Wasn’t scared of AI philosophically, just hadn’t done it. Adding hundreds of sales reps this year without asking what they could do with a fraction of that headcount and a deployed AI BDR instead.

I wanted to cry. Years of their life being left on the table.

I also saw the inverse this week, and it’s a useful cautionary tale for the AI-native side. A $6 billion AI company — I won’t name it — whose agent was quoting our team incorrect pricing. Telling us we’d need to quadruple our spend. Why? No one had properly trained the agent on pricing. When we asked how long the product had been in beta: a year. A year of a badly trained agent making bad calls, at a company whose entire product is AI.

That is what it looks like when smart people don’t treat deployment as seriously as they treat building.

The Companies That Feel Like Summer vs. The Ones You Can Smell Declining

There’s a version of everything I’m saying that’s abstract. Here’s the concrete version.

Walk into a room with a company that’s genuinely operating in the AI era — where the team is deploying tools constantly, where the management is fluent, where AI is in the workflow and not just the pitch deck — and you can feel it. There’s energy. There’s momentum.

NVIDIA right now is the clearest example of summer at a company. Everything firing simultaneously.

Walk into the room of a company that isn’t moving — where the management team is still talking about AI in the future tense, where no one has deployed anything that really moves the needle, where the attitude is “we’re watching this space closely” — and you can smell the decline from the door.

The gap between those two rooms is not technical skill. It’s urgency and willingness to learn.

We’re In The Third Era of AI. The Excuses for Generalists Are Over.

We are in the third era of AI — the generalist era — and the game has changed.

You no longer need to be an engineer to win. You need to be someone who treats AI deployment as a core job function, not a side project. Someone who can answer the question “what did you deploy this month” with a specific, detailed, honest answer.

That is the agentic deployment expert. That is the only hire that compounds right now.

The people who figure this out in the next six months are going to look like heroes inside their organizations. The people who don’t — who keep waiting, who keep saying they’re watching the space, who haven’t trained a single agent — are going to get left behind.

Not because the technology is hard. Because they chose not to learn it when learning was still optional.