What Does It Mean to Be a 'Manager of Intelligence' at Work?
AI is becoming a third economic input alongside labor and capital. Here's what it means to manage it as an individual employee.
What is a “manager of intelligence”?
A manager of intelligence is an employee, often an individual contributor, who decides when, where, and how to apply AI to get real work done, rather than just using AI as a tool someone else configured. The idea comes from a simple observation: companies have always had two economic inputs, labor and capital. AI is now a third. Someone has to decide how it gets allocated, and increasingly that someone is not an executive with a budget. It’s the frontline engineer, support rep, or analyst deciding in the moment whether to let an agent run, where to dig deeper, and when to hand a problem to a human.
TL;DR
- AI now functions as a third economic input alongside people and capital, and nobody has fully figured out how to value or allocate it yet.
- Token and seat based pricing meters compute, not usefulness, which leaves companies unsure whether they’re actually getting a return on what they buy.
- Individual contributors are becoming the real allocators of AI, deciding case by case when an agent should keep working and when a human needs to step in.
- Forward deployed engineering has emerged as a visible version of this role, with engineers from AI labs embedding inside customer organizations to turn raw model capability into working systems.
- Deeper partnerships, including equity stakes and joint ventures, are starting to replace simple software contracts when the value an AI supplier can open up goes far beyond what token pricing can capture.
- Motivation and trust matter more than ever, because unlike a budget or a headcount a manager can direct, employees choose for themselves how seriously to lean into AI.
Remy doesn't write the code. It manages the agents who do.
Remy runs the project. The specialists do the work. You work with the PM, not the implementers.
Why is enterprise AI pricing part of this story?
Procurement teams have spent the past couple of years trying to fit AI into familiar categories: per-seat licensing like traditional SaaS, or per-token billing that meters compute usage. Both approaches answer “how do we pay for this” without answering “how do we know we’re getting value from this.”
Tokens measure how much text a model processes, not whether the output was useful. A company can buy a lot of tokens and still have no idea whether that spend translated into better outcomes, because usefulness depends entirely on how employees apply the model, not on the raw volume of text it generates. That gap, between metered compute and actual business value, is why finance leaders struggle to size AI budgets with confidence. It’s also why so many AI pilots look impressive in a demo and then stall when it’s time to scale.
The deeper issue: a model can be purchased before an organization has learned which work to assign it, how to restructure processes around it, or even how to judge whether its output is good. That know-how doesn’t come bundled with the subscription. It has to be built, and the people building it are usually not in procurement.
Why are individual contributors the ones managing AI?
Capital and labor are both inputs that a manager can direct from above. An officer of a company can allocate budget or assign a team and expect people to execute. AI doesn’t work that way. It’s distributed directly into employees’ hands, and the decisions about how to use it happen constantly, at the level of individual tasks, not in a boardroom.
Consider a routine report. An employee can ask an agent to simply produce it faster, which is a modest efficiency gain. Or the employee can ask the agent to investigate the root cause of a recurring problem the report keeps surfacing, which can create entirely new value for the company. That difference, between using AI to speed up existing work and using it to find and fix underlying issues, is the difference between being a user of AI and being a manager of it.
A concrete version of this: an engineer at an equipment company notices a failure customers keep reporting. An AI agent can pull together maintenance records and compare past cases, but the engineer has to know which operating conditions matter, which records are unreliable, and when a pattern that looks like one fault is actually two different problems described in similar language. The agent can’t make that judgment call alone. The engineer decides when to keep digging with AI and when to stop and bring in a human, case by case. Done across enough cases, that judgment can surface a pattern the whole team can act on, sometimes turning into a new service the company can sell, like ongoing advice that helps customers prevent failures before they happen.
Seven tools to build an app. Or just Remy.
Editor, preview, AI agents, deploy — all in one tab. Nothing to install.
None of that shows up in a procurement contract. It shows up in how empowered and motivated the people closest to the work feel about experimenting with the tool they’ve been handed.
How does forward deployed engineering fit in?
Forward deployed engineering is one of the clearest real-world examples of what managing intelligence looks like in practice. These are engineers, often employed by the AI supplier rather than the customer, who work inside or alongside a client’s organization to turn a general-purpose model into a system that actually functions in that business’s workflows and data.
OpenAI’s deployment arm describes its engagements this way: engineers work with business leaders and frontline teams to identify where AI can help, then design, build, test, and deploy systems tied to the customer’s actual data and processes. Anthropic uses similar language for its own forward deployed engineering work, including a joint venture with firms like Blackstone, Hellman & Friedman, and Goldman Sachs aimed at applying AI inside mid-size manufacturers and regional healthcare systems.
The appeal of this model is that it keeps the “where does intelligence create value” question close to the people who understand both the technology and the business problem. The risk, if it never evolves past a pure services arrangement, is that the client never actually learns to manage the intelligence themselves, they just keep paying someone else to do it. The useful version sits between pure consulting and a packaged product: engineers embedded long enough to transfer real know-how, not just deliver a one-off build.
Are outcome-based deals and equity stakes becoming more common?
Some companies are moving past standard licensing entirely. Sierra, an AI agent company, has been exploring outcome-based pricing, charging based on results delivered rather than tokens consumed or seats licensed. A healthcare payments company called Cedar runs an AI agent (Cora) that handles patient billing calls, including explaining charges and walking people through payment, around the clock. The hospital’s concern isn’t how many tokens that call used, it’s whether the bill got resolved. That’s an outcome, and some vendors are starting to price accordingly.
At the more intensive end, OpenAI has taken an ownership stake in Thrive Holdings, embedding research, product, and engineering teams directly inside Thrive’s portfolio companies, with early work focused on accounting and IT services. That’s a materially different relationship than a software contract. It signals that when an AI supplier believes it can open up value well beyond what token pricing captures, both sides sometimes find it worth sharing more of the upside and the risk through a joint venture or equity arrangement instead.
Outcome pricing is still hard to measure, and nobody should expect every enterprise AI deal to look like this soon. But its emergence reflects the same underlying problem as the manager-of-intelligence shift: standard pricing models assume a company already knows how to convert purchased intelligence into results, and most don’t yet.
Is this actually a new kind of job, or just a mindset shift?
Remy is new. The platform isn't.
Remy is the latest expression of years of platform work. Not a hastily wrapped LLM.
So far it looks more like a mindset shift layered onto existing roles than a formal new job title. The people doing this work are still engineers, support agents, analysts, and account managers. What’s changing is the expectation that they’ll make allocation decisions about AI the way a manager once made allocation decisions about budget or headcount. Companies that recognize this and build motivation, trust, and training around it are positioned to get more out of their AI investment than companies that treat AI purely as a procurement line item.
Frequently Asked Questions
What does “manager of intelligence” mean in the context of AI and jobs?
It describes an employee, often without any formal management title, who decides how and when AI gets applied to a task: when to let an agent handle something fully, when to investigate further, and when to bring in a human. It’s a role defined by judgment over AI allocation rather than by headcount.
Why don’t token-based or per-seat pricing models capture AI’s real value?
Because they meter volume (tokens processed, seats licensed) rather than usefulness. A company can spend heavily on either and still not know if that spend produced value, since the actual return depends on how employees apply the model to specific problems, something no pricing model measures directly.
What is forward deployed engineering?
It’s a role in which engineers, typically employed by an AI lab or vendor, embed inside a client’s organization to build and deploy AI systems tailored to that company’s data and workflows. OpenAI and Anthropic both run programs described in these terms, working with businesses ranging from manufacturers to healthcare systems.
Is outcome-based pricing replacing token and seat pricing?
Not broadly, at least not yet. Outcome pricing is being explored by some vendors, such as Sierra, and is easier to justify in cases like automated customer support where a clear result (a resolved bill, a handled call) can be measured. Token and seat pricing remain the default for most enterprise AI purchases because outcome pricing is harder to define and measure consistently.
Why does employee motivation matter for AI returns?
Because AI, unlike a budget or a team, is applied by individual employees making their own moment-to-moment decisions, not directed top-down by a manager. If employees don’t feel equipped, trusted, or rewarded for experimenting with AI, the gap between buying intelligence and actually benefiting from it tends to stay wide.
