Augment your people. Don’t try to replace them.

There are two things people mean by ‘augmentation’, and the AI moment has quietly merged them.

The old meaning is staff augmentation: you bring in outside people - a contractor, a fractional operator, an agency - to add capability you don’t want to carry as permanent headcount. The newer meaning is human augmentation: you give the people you already have sharper tools, so each of them does more, and does it better. For most of a career those were separate conversations. They are not anymore. A leader now faces the same question in both cases: where do you add capability, and do you add it as people, as tools, or as both?

The loud answer in the market is ‘replace them with AI’. It is the wrong answer, and an expensive one. Here is the case for the boring one: augment your people, invest in what only they can do, and use models where they genuinely earn their place.

The replacement fantasy is a budgeting error

The pitch to a C-suite is seductive because it is phrased as a cost line. Headcount is expensive and visible; a model is cheap and invisible. Swap one for the other and the spreadsheet improves this quarter.

The trouble is that you rarely replace a role. You replace a task. A person in a revenue team does maybe a dozen distinct things in a week, and a model can do a few of them well, a few of them badly, and most of them not at all. Fire the person and you lose all twelve to save the two or three the model could have taken off their plate anyway. You also lose the thing that does not show up on the org chart: the judgment about which of the twelve mattered this week.

The firms that win the next few years will not be the ones who cut the most people. They will be the ones whose people got the most leverage. That is an augmentation strategy, not a replacement one, and it is a choice you make on purpose.

What the research actually shows

The evidence is more interesting than either the hype or the panic. Two findings are worth a board’s attention.

The first comes from a large field study of customer-support agents given a generative AI assistant, run by researchers including Erik Brynjolfsson. The gains were real, but they were not evenly spread. The biggest improvement went to the newest and lowest-performing agents - the tool pulled the bottom of the distribution up toward the top, spreading the tacit know-how of the best people to everyone else. The stars gained little. Augmentation, in other words, is a way to raise your floor, not just your ceiling.

The second is a Harvard and BCG study of management consultants using a frontier model, sometimes called the ‘jagged frontier’ work. On tasks that sat inside the model’s competence, augmented consultants were faster and produced better work. On tasks that sat just outside it, the ones who leaned on the model did worse than the ones who did not - because they had quietly outsourced a judgment the model was not equipped to make, and did not notice. The lesson is not ‘use AI’ or ‘don’t’. It is that the value lands entirely on knowing which tasks are which, and that knowing is a human skill you have to keep investing in.

Put the two together and you get the whole strategy on a postcard: use models to lift the routine work and level up your newer people, and protect - do not automate away - the judgment that decides what is routine in the first place.

What augmentation looks like in a revenue engine

Abstractions are cheap, so here is the concrete version, in the systems we actually build. The pattern is always the same shape: the model does the first draft or the first pass, and a person owns the decision and the outcome.

• Enrichment and research. A model assembles a company’s profile from scattered signals in minutes instead of an afternoon. A person decides whether the read is right and whether the account is worth the team’s time. The model does the digging; the human does the deciding. (More on doing this without the guessing: CRM enrichment without the hallucinations.)
• Lead qualification. A model reads the record and writes a plain-English rationale for why a lead looks strong or weak. A rep reads the rationale and makes the call. Nobody trusts a mystery score; everybody can check a sentence. (How we build that.)
• Notes, summaries and drafts. A model turns a messy call into structured CRM fields and a first-draft follow-up. The person edits it into something true and sends it under their own name. The tedious 80% is gone; the 20% that needs a human is where the human now spends their time.

In every case the person is not doing less thinking. They are doing less typing, and more of the thinking only they can do.

The models, and where the human stays

The tools themselves are good at a narrow, valuable band of work: reading and summarising language, drafting, classifying, and reasoning over text you give them. Anthropic’s Claude, OpenAI’s ChatGPT and Google’s Gemini are all, at heart, extraordinary readers and writers of text. That is genuinely useful in a revenue team, where a huge amount of the work is turning one kind of text - a thread, a transcript, a form - into another kind - a summary, a score, a next step.

What they are not is accountable. A model cannot be the person a customer trusts, cannot carry the context that never got written down, cannot own the consequence of being wrong, and cannot decide what actually matters when two goals conflict. Those are not gaps that a bigger model closes. They are a different category of work. The augmented team keeps all of it firmly on the human side of the line.

Why AI cannot replace your people - the reasoning, not the slogan

‘AI won’t replace you’ is usually said as a comfort. It is worth saying as an argument instead, because the argument is what tells you where to invest.

Judgment under ambiguity. Most real decisions are made with missing information and competing goals. A model can tell you what is typical. It cannot tell you what is right for this customer, this quarter, this relationship - because ‘right’ depends on things that were never in the training data and never will be.

Accountability. Someone has to be answerable when it goes wrong. You cannot fire a model, cannot promote it, cannot ask it to explain itself to a board and mean it. Accountability is not a feature; it is a person choosing to stand behind an outcome.

Unwritten context. The most valuable thing your best operator knows is not in any document. It is a feel for which deals are real, which numbers are lies, which colleague to call. That knowledge is built by doing the work, and it decays the moment you stop letting people do it.

Relationships and taste. Buyers buy from people. Judgment about what is good - a good message, a good hire, a good bet - is a human faculty that improves with practice and atrophies without it. Outsource it wholesale and you do not save money; you slowly lose the ability to tell good from bad.

This is the same line we draw in our AI manifesto: use AI where it beats a rule, keep a human on the decisions that carry consequences, and never let the tool write the things nobody checked.

What a leader should actually do

Invest in both, on purpose. Give your people the tools that take the routine off their plate, and put the time you free up back into the human work - judgment, relationships, the hard calls - rather than pocketing it as a headcount cut. Level up your newer people deliberately, because that is where the research says the leverage is largest. And draw the line, out loud, on what stays human, so nobody has to guess whether the model is allowed to make a decision it should not.

Staff augmentation and human augmentation were always the same instinct: add capability where you need it, without pretending the people are the problem. AI does not change that instinct. It just raises the stakes on getting it right. Bet on your people and give them better tools. The firms that do will quietly pull ahead of the ones still trying to subtract their way to a better number.

Common questions