20 People. $100M Revenue. The 5 Operations Behind Every Tiny Team Beating a Giant One.

Every workforce skill in history so far has had a finish line, a point where it was done.

AI doesn't, picture a bubble, the air inside is everything AI agents can do reliably today.

The air outside is everything that still requires a person.

The surface of that bubble, that thin curved membrane, that is where the interesting work is happening right now.

It's where you decide what to delegate and what to keep, how to verify agent output, where to intervene, how to structure that handoff, working on that surface well is the most valuable professional capability in the economy today.

But here's the thing, that bubble is inflating.

Every model release, every capability jump, every quarterly leap in reasoning or context or tool use, the bubble keeps getting bigger, tasks that sat on the surface migrate inside the bubble where agents are and the boundary continues to shift out.

And a person who learned to work on the surface of the November bubble is now standing inside it doing work that an agent handles better than she does, running verification checks against failure modes that don't exist for AI models anymore.

But notice what else happens.

But that's the simple part of the story.

Everyone's talking about that.

Let's go beyond it.

Notice what happens when a bubble expands.

The surface area increases.

The frontier doesn't shrink as AI gets more capable.

It actually grows.

There's more boundary to operate at, more places for human judgment, not less.

More seams between human and agent work, not less.

More judgment calls about what crosses that AI agent membrane and what doesn't.

More verification challenges at this new edge, more decisions about where human attention creates value that it didn't need to create before.

Every prior workforce skill, whether you're talking about literacy or numeracy or computer literacy or coding, was a destination.

You reached it, you got it, you're done.

The target doesn't move.

But the skill of working at the surface of this bubble in AI has no fixed destination because the surface is always expanding outward.

You can't learn it once.

You can learn to stay on it, to move with it as it expands.

To maintain your footing as the curvature shifts beneath you.

But that's a fundamentally different kind of skill than any workforce development system has ever been asked to produce.

We are trying to teach this expanding surface skill set mostly with fixed destination methods.

Every curriculum, every certification, every training program has assumed the target stands still.

This one doesn't, and the mismatch between the skill the economy needs and the infrastructure we've built is the most expensive gap I've seen in the global workforce.

And I'm giving that skill a name.

It's called frontier operations.

The surface of that bubble is the frontier.

Working on it, sensing where it sits, structuring handoffs between AI and human across it, maintaining a model of how agents fail at the current edge.

Forecasting where the surface expands next.

Deciding where your attention creates the most value.

That is the frontier operation.

That is frontier AI operations at its best.

This is not AI literacy.

That's just knowing what a language model is and how to write a prompt.

It's the equivalent of teaching someone the alphabet and then calling them a reader.

It's also not prompt engineering.

That's one technique inside one component of the practice.

It's like calling surgery scalpel handling.

And it's also not that vague gesture at human judgment that tends to fill a lot of keynotes about the future of work.

Because most people correctly identify that judgment matters, but they incorrectly assume that naming it is the same as teaching it in this fast-moving world.

Frontier operations is the specific, practicable, accessible version of the skill everybody is pointing to.

And very few people are building.

It does have components.

It does develop through practice.

It degrades if you don't maintain it.

And it's the first workforce skill in history that expires on a roughly quarterly cycle.

Which means everything we know about how to teach workforce skills doesn't work too well for this skill.

So what's inside the box?

I've named it as frontier operations.

But I'm not just going to point at it and stand there vaguely and tell you it's great.

There are five kinds of frontier operations.

If you picture the world as an expanding bubble of AI capability, which I think is very accurate.

There are five kinds of skills that stay persistent across the expanding surface area of that bubble.

The first one is what I call boundary sensing.

It's the ability to maintain accurate, up-to-date operational intuition.

About where the human agent boundaries sits for a given domain.

This is not static knowledge.

It updates with every single model release, every capability jump, every subtle shift in how agents handle long context or tool use.

Opus 4.5 couldn't reliably retrieve information from deep in a long document.

Three months later, Opus 4.6 scores 93% on retrieval at 256,000 tokens.

A person who calibrated his boundary sense against the November model and hasn't updated that is now either overtrusting or underusing the February model.

And both kinds of errors are very expensive.

So the skill is maintaining the calibration, not having it once.

So what does this look like in practice?

For a product manager, good boundary sensing looks like letting an agent draft a credible competitive analysis.

But realizing that that same agent will miss the political dynamics between two executives that it's never observed or been given context on.

And so the product manager doesn't hand the whole analysis to the agent.

They hand the agent the market sizing and the feature comparison piece.

And those are tasks that are now safely inside the AI bubble.

And then the product manager reserves the stakeholder dynamic section for themselves.

And now last quarter, that bubble was smaller and maybe the feature comparison was updated at least halfway manually.

Not anymore.

It's updated.

Let's say you're a marketing director, that might look like an agent producing a solid first draft campaign copy and AB test headline variants.

But maybe brand voice copy drifts subtly off tone after the third or fourth version.

Well, you use an agent for ideation, you use an agent for first drafts, you edit the voice yourself.

And you don't ask for it to iterate past version two.

What bad would look like here is the marketing director trusting everything and getting burned by hallucinations.

Or trusting nothing and doing everything manually.

Most commonly, the marketing director probably calibrated six months ago and hasn't noticed the boundary move.

The marketing director doesn't have boundary sensing as a skill set.

Here's the second skill set.

Seam design.

It's the ability to structure work so that transitions between human and agent phases are clean.

Verifiable and recoverable.

This is very much an architectural skill.

It's closer to how a good engineering manager thinks about system boundaries than to how an individual contributor thinks about their tasks.

The person doing seam design is asking, if I break this project into seven different phases.

Which three are fully agent executable?

Which two need human in the loop?

And which two are still irreducibly human?

What artifacts pass between each of these phases?

What do I need to see at each transition to know that things are on track?

The reason this is a very distinct skill and not just project management in a trench coat is that the answer changes as capabilities shift.

So the seam that was in the right place last quarter is in the wrong place this quarter.

And so the skill isn't in the design one off.

It's in the ability to redesign as agent capabilities.

So this might look like a software engineering lead at a company structuring the handoff so that ticket triage and work routing go to the agent.

Architectural decisions stay with humans.

And the boundary between those two is then defined by specific artifacts.

The content of the ticket, the structure of the code base.

The org chart, and then you have specific verification checks at that scene that ensure the handoff is clean.

Without those, you either go end to end with agent runs and you don't have the verification infrastructure in place yet to do that reliably.

Or you have humans manually reviewing things the agent now handles better than they do.

Another one might be a consulting engagement manager.

Let's say they break the strategy project into research, which is agent led with a human defined scope, and synthesis, which is human led with agent generated first pass frameworks, and a client presentation, which is human led with agent generated slide drafts.

The seam between research and synthesis is a structured deliverable.

It's a fact base with source citations that human can spot check in just a few minutes.

Months ago, that seam would have included manual fact verification on every data point, but the agent citation accuracy has improved dramatically, and so it made sense for the consulting engagement manager to move the scene.

The third skill is failure model maintenance.

It's the ability to maintain an accurate, current mental model of how agents fail.

Not that they fail, but the specific texture and shape of failure at the current capability level.

That's really important.

Early language models failed very obviously, right?

They had garbled text, they might have wrong facts.

They might have incoherent reasoning.

Current frontier models fail in very subtle ways, they have correct sounding analysis built on a misunderstood premise.

Or plausible code that handles the happy path and breaks on edge cases.

Or research summaries that are 98% accurate while the remaining 2% are confidently fabricated in a way that's very difficult to distinguish from the accurate parts unless you know the domain well.

So the skill here is not be skeptical of AI output.

That's necessary, but not particularly useful.

It's like saying the skill of surgery is to be careful.

The skill is actually maintaining a differentiated failure model, knowing that for task type A, the agent's failure mode is X, and here's how to check for it well in your domain.

While for task type B, the failure mode is Y, and and there's a different check for it.

This might look like a corporate counsel that knows an agent reviewing contracts is going to catch boilerplate issues, but miss indemnification clauses.

Or maybe miss non-standard termination language and maybe miss the interaction between a specific liability cap and section seven and a carve out varied in the exhibit.

So the failure model right now might say, trust the boilerplate scan, manually review cross references between liability provisions and exhibits.

And that's that's a very different check than read the whole thing again.

And it takes much less time and it allows you to be much more precise in your work and much more efficient.

A data scientist might know that an agent generating Python for data analysis will handle Pandas transformations and standard statistical tests very reliably.

But produce somewhat plausible sounding nonsense when the data has messy edge cases, like mixed data formats or implicit nulls or columns that changed meanings mid data set.

And so the failure model says, verify the data cleaning steps and the assumptions about column semantics.

And then trust a lot of the downstream analysis if the cleaning is correct.

And what does bad look like here?

Bad would look like applying the same generic skepticism to everything, which is very inefficient.

Right, it assumes memorized failure patterns from six months ago hold when that's actually incorrect.

Here's the fourth skill.

Capability forecasting.

It's the ability to make very reasonable short-term predictions about where that bubble boundary for AI will move next.

And to invest learning and workflow development accordingly.

This is not about predicting the future of AI.

It's actually something that nobody could do super reliably over long horizons.

It's about reading the trajectory well enough to make very sensible six to 12 month bets about what is likely to become agent territory.

Think of it like reading the swells on the ocean.

A surfer doesn't predict exactly what the next wave will look like, but a good surfer will read the sea.

Understand how the floor shapes wave at this particular break.

And position themselves where the next rideable wave is most likely to form.

And that skill is very much probabilistic positioning.

Not linear prediction.

So a person with good capability forecasting in early 2025 could look at the trajectory of coding agents, 30 minutes of sustained autonomy.

And then seeing how it scales from there and start investing more in code review and specification skills rather than just raw coding.

Meanwhile, a UX researcher watching agents get better at survey design and qualitative coding can start investing in interpretive synthesis, the skill of turning coded data into product insights that will shift a road map.

So the coding is migrating inside the bubble.

The so what of the coding is where the new surface happens to be.

So what does bad look like here?

It can look like chasing every new tool, which is exhausting and doesn't really generate compound returns.

Or it can look like ignoring developments until you're forced to catch up.

Most commonly, it tends to look like investing heavily in learning a particular platform, and then that platform's advantage might evaporate when the next model ships.

And either deletes that part of the workflow or provides leverage that that platform can't deliver on its own.

We've seen this over and over and over again with AI.

As large language models eat up more and more and more of the capability space.

And skill number five is leverage calibration.

It's the ability to make very high quality decisions about where to spend human attention, which is now the scarcest resource in an agent rich environment.

As agent capabilities continue to increase, the bottleneck shifts from getting things done to knowing what things are worth a human's attention.

Even McKinsey has a framework that describes two to five humans supervising 50 to 100 agents running an end-to-end process.

And that's not just McKinsey, that's a lot of different agentic patterns coalescing across the industry.

That roughly 10 to one ratio makes the math of attention very, very clear.

If you have 100 streams of agent output and eight hours a day, you cannot review everything at the same depth.

The skill is going to be triaging your own attention in real time.

So this might look like an engineering manager overseeing agent assisted development across five teams, and that person develops a hierarchical attention allocation.

Most agent generated code flows through automated test suites and gets linted on its own.

A smaller subset, like billing or data pipelines, might get flagged for human code review.

Only architectural decisions and cross-system changes might get flagged out for deep human engagement.

And so a good manager would recalibrate those thresholds monthly because the agents keep getting better at the routine tier.

And you keep finding new categories that belong in that middle in between like human touch tier.

Ahead of customer success overseeing agent handled support tickets.

Might review escalations and a random sample of resolved tickets to ensure quality.

But she wouldn't review the resolution of routine password resets.

She does review every ticket where the agent access account modification tools.

And that threshold ends up being calibrated to risk and adjusted as the agent's ability to use tools on her particular ticket system get better and better.

Now, what might bad look like here?

It might look like reviewing everything at the same depth, which creates a bottleneck.

Which creates human burnout.

Or it might look like reviewing nothing.

Which you would only do if you were intentionally piloting a dark factory floor scenario.

And that is something that very, very few teams are ready for.

Even if technically speaking it's possible today.

These are not a checklist.

These are five operations that are simultaneous, integrated and continuous.

It's a it's a way of practice, like driving involves steering and speed management and route awareness and hazard perception, all at the same time.

At any given moment, a person operating at the frontier is sensing the current boundary, they're designing seams around it, they're verifying against an updated failure model.

And they're making bets about where that boundary is going to move.

And allocating attention across that system.

The integration is what makes this a practice and not a curriculum.

You can teach each component in isolation, but a person who's good at all five individually.

And doesn't run them simultaneously, still isn't operating at the frontier.

Because they're thinking about putting those skills into practice.

Instead of doing them together seamlessly.

As a way of business.

As the way of working.

As the future of AI human collaboration.

This skill gap is structural.

Every other AI adjacent skill might eventually get absorbed into the technology itself.

Prompting techniques are starting to get baked into system defaults and prompting is moving up a level into things like intent engineering.

Integration patterns are getting productized.

But frontier AI operations, you can't really automate them because by definition, there's a surface of AI capability.

And so when a task migrates inside the AI bubble, the surface just expands outward.

And the person who operates at the surface moves with it.

The skill is sort of structurally resistant to its own obsolescence.

I will also call out that the structural gap compounds.

A person who develops this skill set six months sooner than their peers, they don't just have a six-month head start.

They have six months of updated calibration that the peer doesn't have.

And because capabilities are accelerating, the distance between calibrated and uncalibrated keeps getting wider with every model release.

And so the person whose boundary sense was current in February and the person whose boundary sense was current last August are operating worlds apart.

This is really the mechanism behind those gigantic leverage numbers that keep appearing in production deployments.

When you see, you know, cursor hitting stunning numbers with a small number of people from a revenue perspective, lovable doing the same thing, when you see the Anthropic team shipping every time I turn around.

The gap between shipping and revenue at AI native companies and shipping and revenue at traditional SAS companies isn't really explained by better tools.

It's explained by people who have developed the operational practice to be on the bubble.

And convert those tools into reliable output as AI continues to evolve.

I believe that this skill set is the single largest determinant of not only which businesses tend to succeed over the next decade.

But which economy starts to win over the next decade.

It's not going to just be about which economies built models.

Because models are portable over the internet.

It's not just going to be about which economies have compute.

Because compute can be rented.

It's going to be about which economies can field workers who are excellent at operating at the AI human frontier.

The models are starting to commoditize, the compute is being built out everywhere.

What is going to continue to be scarce is the human capacity to convert those inputs into economic output.

And that conversation is going to be something that we need to have a lot more of.

As compute starts to become more abundant.

So you might wonder, if you're trying to foster this skill set.

What does that actually look like?

Well, one, as a leader, you should be building practice environments, not coursework.

Just like flight simulation helps you learn to fly.

Simulating AI environments in sandboxes is important for AI era work.

This is much, much more important.

Practice environments are where agents have different capability levels, where failure modes are realistic.

Where rules can change so practitioners have to recalibrate.

This is much more practical.

Than just looking at a bunch of slides and saying you did an AI workshop.

You have to get in and touch the AI often in order to get skilled.

Measure calibration, not knowledge, is the second thing I would call out for leaders.

The right assessment is not, you know, can you write a good prompt?

It is given a task and an agent at capability level X, can you accurately predict where the agent is going to succeed?

Where it's going to fail?

And how to structure your work accordingly?

That is a much harder thing to do than writing a good prompt.

But it's a much, much more durable skill because it measures the ability of humans to work with AI as AI continues to scale.

Third, maximize feedback density, not training hours.

The speed of skill development is really a function of how many cycles a person gets through with AI per unit of time.

And it's not really a linear function of hours of training.

So a person that completes a 40-hour AI course offsite and then returns to the workforce, but never really touches an AI tool beyond some light chat GPT.

Has zero calibration cycles.

A person who skips that course and then delegates 10 real tasks a day to an agent and evaluates the output.

Is going to have 100 of them in 10 days.

Make sure that you understand how AI exposure is actually shaping feedback density for your team.

Fourth, create explicit roles for frontier operations.

The skill doesn't develop if it's just an undifferentiated part of somebody else's job.

Organizations need people whose specific function is to operate at the boundary, to maintain the failure models.

To update verification protocols, to redesign workflows when capabilities shift.

You can call them AI automation leads.

You can call them delegation architects.

You can call them frontier engineers.

The title matters less than recognizing that this process of evolving the automation frontier is very high leverage.

And it's a distinct specialty and it's something that you need someone to focus on.

So that you can aggressively socialize changes through the business.

A lot of the org chart built for the pre-agent era assumes that output scales with head count.

More people equals more output.

But with frontier operations, you're inverting this.

Output scales with leverage and leverage scales with how well a small number of humans operate at that boundary.

So the organizational unit that matters is like a tiny pod.

And I've seen two structures emerging here that are significant.

Right, the first is a team of one, a single person with a very strong frontier operation skill set.

Who runs multiple agent workflows across a domain.

That person does the boundary sensing, defines the seams, maintains the failure models.

And calibrates attention allocation for the pod.

And others are executing, don't worry, with lots of AI within those structures.

And developing their own frontier intuition as AI models evolve through practice.

This is a little bit like how a surgical team operates.

One lead that sees the whole field, others separated by complementary skill boundaries.

Executing in roles that mesh well together.

So in product development, this might look like one frontier operator who owns the human agent workflow across the product surface.

A couple of engineers executing heavy agent assisted development.

Maybe a designer running agent assisted prototyping and user research and also committing code.

And maybe a data scientist managing the analytics pipeline.

They ship at the pace of a 20-person team because the operator keeps the seams really current.

And the failure modes calibrated.

And oh, by the way, the operator is shipping too.

And the rest of the pod has enough frontier skill to execute without supervision.

So you might wonder, how does this ladder up?

I think the simplest way to think about this is to think about where you want to double down on your bets.

If teams of one are able to effectively discover opportunities for deeper engagement, and teams of five are able to build meaningful products.

A team that ladders up from there is really focused on one of two areas.

Either you have a portfolio of bets you're managing as a leader and you're farming that out across four or five teams of five.

Or you're in a position where you're looking for one big win and you pick something.

From a team of five or a team of one that's doing exploratory work in your area, and then you rally your whole team to double down on that big bet and produce something really polished.

And and how you allocate that depends a lot on your business strategy and product strategy.

Which has to devolve much farther down from executive leadership than it used to.

You need people who are managing four or five teams of five to be just as strategically informed as your CEO at this point.

So what do you look for when you're hiring for this skill set?

It's not necessarily a case for traditional hiring signals, right?

Credentials, years of experience, tool proficiency.

Those aren't necessarily helpful.

Instead, you want to look for, does this person track where agents succeed and fail in her domain?

Can she articulate specifically what an agent handles today and where it doesn't handle well?

Can she describe a new capability or immediately start redesigning a workflow?

Or does something just get filed under interesting and never get action?

Does this person have a failure model?

Not generic skepticism, but a differentiated understanding of how agents fail on which tasks.

Is there a reliable trend of forecasting this person has maintained where you can see them?

They've got really good instincts about where the future is going.

The person who can answer these questions well at high quality.

That's your frontier operator.

The person who answers them with, well, I'm good at prompting.

That's not your frontier operator.

Let's say you want to get better at this skill.

What does that mean for you?

If you're an individual contributor, you need to start tracking where your boundary sense is incorrect.

You need to track where agents surprise you.

The surprise is a signal.

You want to be collecting your surprises on purpose, logging them and starting to build your professional instincts.

So that you start to understand better where agents operate, where agents work, where agents don't work.

And if your agent hasn't surprised you recently.

Then you are definitely not operating at the boundary.

If you manage people, you want to look at how your team allocates attention across agent assisted work.

Are they reviewing everything at the same depth?

Is there a bottleneck that's masquerading as due diligence?

Are they reviewing nothing?

The right answer is going to be differentiated based on your domain.

But it's up to you to figure out where human attention ought to be allocated here.

And if your team can't articulate to you their philosophy of human attention.

You've got a problem.

If you run an organization, the question isn't, are we using AI?

It's, do we have people whose job it is to know where the evolving AI agent human boundary is and how we think about redesigning our workflows as it shifts?

If the answer is you can't name someone.

You are leaving one of the most consequential organizational capability decisions.

Of the decade to chance.

And I wouldn't do that.

We have seen context windows.

We have seen retrieval.

It is hard to overstate how much models have gained in capability between November of 2025 and February of 2026.

So within the last 60 to 90 days.

Anyone who is deep in AI has felt the difference.

And you feel it whether you're touching Opus 4.6 versus 4.5.

You feel it touching Codex 5.3.

You feel it touching Gemini 3.1 Pro.

And if you are listening to this and you can't feel it, if you're like, I don't know what the difference is.

That means you're not at the edge of the bubble.

And that was just one quarter, the next quarter, there's going to be another shift.

Everything we are seeing out of the labs is indicating that we are not slowing down.

And so if you are not in a place where these February agents are surprising you.

The best thing you can do to welcome yourself to the frontier is to find a way to give your agents a job that surprises you.

Whether they fail, whether they partly succeed.

Give them something that allows them to surprise you.

Because if you don't, you run the risk of missing this expanding capability bubble.

And this is the new workforce skill set that will define all of our career success for the next decade.