AI is going to reduce headcount.
That seems to be one of the few things everybody has already decided about it. If one person can get more done, a company needs fewer people to produce the same amount. The maths appears straightforward.
Businesses are starting to act on it.
Some aren't replacing people when they leave. Others are making redundancies or building systems specifically to remove work that used to need a person.
This is one possible response to AI. But it isn't the only one.
The same tool, two decisions
Imagine two software companies, each with twelve engineers.
At both, AI has made the work quick enough that nine engineers can now produce roughly what twelve did before.
The first has a mature product and wants to reduce costs. It cuts the team to nine and keeps its plans roughly where they were. The same amount gets done with a smaller salary bill.
The second has customers waiting, new products it wants to build and more useful work than its team can reach. It keeps all twelve engineers and expands its plans. If the opportunity is large enough, it hires more.
Both decisions can make sense. One company uses the gain to spend less, whilst the other uses it to attempt more.
The technology is doing the same thing in both places. The difference is what the business needs from it.
This is why broad predictions about AI and headcount aren't very useful. They treat every company as though it has the same problems, the same opportunities and the same appetite for growth.
AI changes what a team can do. It doesn't decide what the company should do next.
The obvious option
Reducing headcount is the easiest benefit to put into a spreadsheet.
Salaries are a large cost. If AI removes enough work, employing fewer people can improve the numbers quickly and visibly. Not replacing somebody who leaves is quieter than a round of redundancies, but the result is similar. The team becomes smaller and the business expects the tools to close the gap.
Sometimes that will work.
Plenty of jobs contain repetitive work that software or AI can handle faster and more consistently. Some roles will shrink, and some will disappear. Pretending otherwise doesn't help the people affected by it.
The risk is assuming that removing tasks is the same as replacing a person.
A job is rarely one repeatable activity. It includes noticing the unusual case, understanding why a customer is unhappy, deciding when the normal process is wrong and taking responsibility when something fails.
AI can remove a large part of the visible work whilst leaving the difficult edges behind. The company saves time on the common cases, but still needs somebody who understands the system well enough to handle everything else.
That may still mean fewer people. It doesn't always mean no people.
The other option
Most software teams don't have a shortage of things they could do.
They have features waiting to be built, old systems that need attention, customers asking for improvements and small problems that have been tolerated for years because something more urgent always won.
Give that team more capacity and it doesn't have to become smaller. It can finally do more of the work that matters.
AI can shorten the distance between an idea and a working version. Engineers can explore several approaches, understand unfamiliar code more quickly and spend less time on repetitive implementation.
The benefit doesn't have to appear as fewer salaries. It can appear as a better product, more experiments, faster learning or work that would never previously have justified the cost.
A company can also hire into that advantage.
If an engineer with good judgement can produce more useful work with AI, that engineer has become more valuable, not less. Employing another one gives the company more of an increasingly productive resource.
This is why renewed demand for software engineers isn't a contradiction. The tools make each capable engineer more productive, but they also make far more software worth building.
Cheaper work creates more work
The amount of software a business wants isn't fixed.
When software is expensive, only the most important ideas survive. Internal tools remain clumsy, manual processes continue and small customer problems aren't worth solving.
As the cost falls, the threshold falls with it.
An idea that couldn't justify three months of a team's time may easily justify a week. A system that was too expensive for one department becomes practical. A small business that couldn't afford custom software can suddenly consider it.
Making software cheaper doesn't only reduce the number of people needed for today's work. It creates demand for work that previously didn't make economic sense.
We tend to imagine a fixed pile of tasks being completed by fewer people. In reality, the pile grows as more things become possible.
The companies that see AI only as a way to reduce the cost of their existing plans may miss the larger opportunity. They'll arrive at the same destination with a smaller team whilst somebody else uses the same tools to travel further.
Replacing people creates engineering work
There is another odd effect in all of this.
Building systems to replace human work is still work.
Those systems need to connect to real products and messy company data. They need boundaries, monitoring and somebody who can tell when the output is confidently wrong. They need to change when the business changes and recover when a service stops behaving as expected.
The more responsibility a company gives to AI, the more important the surrounding engineering becomes.
That doesn't mean the same jobs will remain in the same places. A business may reduce headcount in one function whilst hiring engineers to automate parts of it. The overall number can fall even as demand for particular skills rises.
This is uncomfortable because productivity gains aren't shared neatly. The person whose work is removed isn't necessarily the person who gets the new opportunity.
But it explains how companies can be reducing headcount at the same time that software engineers, particularly those who know how to use these tools well, become more sought after.
Fewer people or a bigger ceiling
AI will reduce headcount in some companies.
People will leave and not be replaced. Roles will disappear. Systems will be built because replacing a process with software is cheaper than continuing to employ people to run it.
Other companies will make a different bet.
They'll keep their teams, hire more capable people and use the extra capacity to build products, enter markets and solve problems that were previously out of reach.
The interesting question isn't how many people AI allows a company to remove.
It's what the company believes it could achieve if every person became more capable.
One answer produces a smaller team.
The other produces a bigger ceiling.
