What Companies Need Isn't an FDE—It's a TRC: Compute Deserves a Manager

There’s a role the AI industry can’t stop talking about right now: FDE, Forward Deployed Engineer—a vendor engineer embedded inside a client company. It isn’t a new role; Palantir has run this play for over a decade. But it came roaring back this year.

Why? In May 2026, OpenAI and Anthropic each founded a company dedicated to exactly this—putting engineers inside other people’s businesses. OpenAI put in $4 billion at a $10 billion valuation; Anthropic teamed up with Blackstone and Goldman Sachs on $1.5 billion. Job postings are up more than sevenfold in a year, and senior people can clear ¥5 million a year.

FDEs are hot—but whose side are they on?

The pitch sounds great: the AI company sends someone to your business to help you get more out of AI. But put plainly, the reason an AI company wants this role inside your walls is to get you consuming AI compute more steadily, more reliably, and in greater volume, so they make more money. Anthropic says as much itself: every $1 of software spend should pull in $6 of services.

And that’s exactly where the problem is. This expensive specialist is paid by the AI company, and that settles it structurally: their loyalty can never truly sit with the company they’re deployed into.

So the thing I do should be called a TRC

For more than a year now I’ve been doing the opposite: burrowing into how a company actually operates and runs its business, undercover, and helping that company use AI better. How is that different from an FDE? A couple of days ago a term came to me, and I think what I do should be called TRC—Token Resources Consultant.

You take the company’s point of view and manage compute and AI tools as a resource portfolio, provisioning the tools and compute that actually fit instead of AI-ifying everything in sight. Here’s the work: selection (screening which AI tools and which models you should be on), custom builds (shaping the right AI tooling around your business), configuration (tuning how it’s used to the optimum), planning (how the whole compute budget gets allocated and spent), and governance (don’t get locked into a platform, don’t leave your data exposed).

One key point: a TRC doesn’t have to be on your headcount. They can be an outside consultant. The distinction isn’t whether they’re on your payroll—it’s whose side they’re on.

Compute is the fourth factor of production, and nobody owns it

More and more companies are pouring real money into AI tools and compute with nobody experienced coordinating any of it. Even mature, sophisticated tech companies. miHoYo talked about this at an Alibaba Cloud summit: one of their engineers wired up a few dozen AI agents to work together, and the agents fell into an infinite loop and spun all night—¥2 million of compute burned in a single evening.

In a company, people are “human resources” and HR manages them. But compute—a production input whose spend keeps climbing—has nobody coordinating it, nobody managing it, and in some places nobody accountable for it at all. I’d go further: companies are going to grow a new department for this, a compute resources department. The ideal end state is that they build that team from the inside. Right now, though, there aren’t nearly enough people with the skill set, which is why so many companies want an experienced outside brain to carry the job first.

It always comes back to the same thing: manage the resources from the company’s side, so AI serves real business instead of some model or tool vendor’s interests. Otherwise you really are just burning money.

Real adoption means the CEO/founder gets in the game themselves

One more thing I strongly agree with: an AI transformation has to be pulled by the person at the top. The CEO, the founder, has to get in the game personally—at minimum they need to know where the boundary of what AI can do in their own business actually sits.

I coach a lot of CEOs and founders on AI now, and the thing that hits me hardest is this: a lot of the AI that actually gets used isn’t something I built for the company. It’s the founder who, while working hands-on with top-tier AI, rebuilds their own business model—sometimes invents an entirely new one. I can’t do that as an AI specialist. Only a founder with deep industry insight, standing on the front line, can. All I do is make it easier for them to get their hands on the best tools, get moving quickly, and hit fewer potholes.

What’s next

Starting today, my WeChat Channels content will focus harder on “how AI serves a company’s real business.” That will probably drive my traffic down, and that’s fine—the community gets more focused. CEOs, founders, industry experts, and big-company executives across industries, colliding every day over their specific AI practices. That alone is exciting enough.

Last thing: I haven’t carved out much time for consulting, which makes me look picky about clients, and I’m sorry to the people who couldn’t get a slot. It isn’t that I’m turning anyone away—time really is limited, and I pick the industries I’m better at and enjoy more. Questions in the comments are welcome too; I still pull episode topics from there.

Source: EP0053_audio.mp3 · ASR model gemini-2.5-pro (chunked parallel) · full text of the original recording

[00:00] Friends, I’ve honestly got way too many projects going right now—I was still answering questions in the group chat at 2 a.m. So today I want to talk about a role the AI industry can’t stop talking about lately: the FDE, the Forward Deployed Engineer. It isn’t a new role, but it came roaring back this year. In May, OpenAI and Anthropic each founded a company to do exactly this—put engineers inside other people’s companies. OpenAI put in $4 billion, at a $10 billion valuation now. Anthropic teamed up with Blackstone and Goldman Sachs on $1.5 billion. Both of them are paying very big salaries

[00:25] to poach talent in this area. Job postings are up more than sevenfold, and the pay runs north of ¥5 million. So what does an FDE actually do? It sounds great when you say it: the AI company comes into your business and helps you get better at using AI. But put plainly, the reason an AI company wants this kind of role inside your walls is to get the company consuming AI compute more consistently, more reliably, and in greater volume, so the AI company makes more money. Anthropic says as much itself: every $1 of software spend should pull in $6 of services. And that’s exactly where the problem is.

[00:50] This highly paid person shows up inside your company, but their loyalty sits with the AI tool. Their salary is paid by the AI company. That alone means this role’s loyalty can never sit with the company. So here’s the question. I’ve been doing something similar for more than a year now—what’s the difference? I also claim to get deep into how a company operates and runs its business, to help the company use AI better, going undercover inside their own business. How is that different from an FDE? A couple of days ago this term suddenly came to me. I think what I’m doing now

[01:16] should be called TRC—the company’s Token Resources Consultant. You take the company’s side of the table and manage compute and AI tools as a resource, provisioning the AI tools and the compute that actually fit the business, instead of AI-ifying everything in sight and instead of just cutting headcount. More and more companies are pouring a lot of money into AI tools and into compute with absolutely nobody experienced helping them coordinate and manage any of it. Even mature, very large tech companies. miHoYo, for instance,

[01:41] talked at an Alibaba Cloud summit about how one of their engineers wired up a few dozen AI agents to work together, and that night the agents fell into an infinite loop and spun for a whole night—¥2 million of compute burned in a single evening. In a company, people are called human resources and there’s an HR function to manage them. But compute—a production input we spend more and more on—has somehow ended up with nobody coordinating it, nobody managing it, nobody even accountable for it. I actually think future companies will have a department called the compute resources department, and companies should grow that team internally. But right now, there just aren’t enough people who can actually manage compute

[02:06] as a resource. So I think right now there’s enormous demand for an experienced outside team to take that job on. I call it the Token Resources Consultant. The core of it is: you sit on the company’s side and run the resource—you handle selection, configuration, planning, and governance of compute resources. You’re not standing on the side of some particular model or AI tool vendor. You’re genuinely on the business side, making AI serve the real business. Otherwise you really are just burning money. The other thing I strongly agree with is

[02:31] that when a company goes through an AI and digital transformation, it has to be led by the CEO or founder themselves. The CEO, the founder, has to get in the game personally—at minimum they need to know where the boundary of what AI can do inside their own business actually is. I’m coaching a lot of CEOs and founders on AI right now, and the thing that hits me hardest is that a lot of AI adoption isn’t something I made happen for the company. It’s the founder, in the process of using the very best AI, restructuring their own business model—sometimes building an entirely new one.

[02:57] And that’s something I, as an AI specialist, have no way of doing. Only a CEO or founder with really deep industry insight, on the front line, can pull that off. What I do is just make it easier for them to get their hands on the best tools, get up to speed fast, and hit fewer potholes. A couple of days ago I also saw a data point estimating Claude’s current daily active users at only 500,000 to 700,000. Between the people who’ve actually got it working and the people who haven’t, there’s a massive gap in information and in ability to process information. You can feel it directly—your ideas get built out fast,

[03:22] and productivity goes up 20x, even 50x or more. Things where the information used to be patchy and the logic wasn’t clear—now you can quantify it, analyze it, and actually make the call. It’s like a brain-computer interface—an evolutionary jump, and nothing else replaces it. So starting today, my WeChat Channels content is going to focus much more on how AI serves a company’s real business. That decision will probably cost my account some traffic. But that’s fine—the community will be more vertical. Cross-industry CEOs, founders, industry experts, and senior execs from the big companies

[03:47] are in there every single day, comparing notes on their concrete practices for using AI in their business. It’s genuinely exciting. And finally, let me say something about the consulting side. I never carved out much time to take consulting work in the first place, so it can look like I’m being picky about clients. To those of you who haven’t been able to book a session, I owe you an apology. It’s not that I think you’re not a good client. Time really is limited, and I pick the industries I’m better at, or that I like more, to support. Please do leave your questions in the comments too. I still pull my topics from the comments and make videos to show you. Thanks again for following.

[04:13] See you next episode, bye bye.