There is a specific kind of headline making the rounds right now, and it follows a pattern. A company announces record earnings. The press release glows. Then, buried in the same quarter, comes the other number: hundreds of support roles eliminated. Workforce reduced. Team right-sized for the new environment.
Cloudflare is the example I keep coming back to. Best quarter in the company's history, by their own account. And 1,100 people let go in the same breath. The subtext is not subtle. The AI can handle the tickets now. The support staff is no longer load-bearing.
I understand the logic. I think it's shortsighted. And I think we have a specific, concrete counterexample that shows exactly what the alternative looks like — and what it produces.
// The IKEA Bet
When IKEA looked at its customer support operation, they saw the same thing Cloudflare saw: a chatbot could handle a large percentage of the volume. Self-service tools, AI-assisted responses, automated resolution flows — the case for replacing the people was straightforward and the cost savings were real.
But IKEA's leadership noticed something else. Their support staff knew their products with a depth and intimacy that is almost impossible to manufacture. These weren't order-takers. They were people who had spent years understanding what customers were actually trying to do, what went wrong, what questions came up again and again, what it looked like when a product succeeded or failed in a real home. They knew the catalog not as SKU numbers but as objects in lived space.
So instead of replacing that knowledge, IKEA redirected it. The support staff were retrained as interior designers — people equipped to help customers not just resolve a problem but actually plan a space. And that pivot didn't just preserve headcount. It opened a new billion-dollar business line.
The AI didn't eliminate the value of those people. It freed them from the rote parts of their job so the irreplaceable part — the deep, contextual, product-intimate understanding — could finally be fully deployed.
// What Jevons Knew
William Stanley Jevons was a nineteenth-century economist who noticed something counterintuitive about coal. As steam engines became more efficient — as the cost of producing useful work from coal dropped — you might expect consumption to go down. More efficiency, less waste, less fuel needed.
Instead, consumption went up. Dramatically. Because efficiency made coal-powered work more economically viable across a much wider range of applications. More industries, more machines, more use cases that had previously been priced out. The lower the cost of the input, the larger the demand for the output.
This is called Jevons' Paradox, and I think it describes exactly what is about to happen to software.
Right now, the narrative is that AI is making software development cheaper. This is true. What follows from that in the standard framing is: fewer engineers needed, lower salaries, smaller teams. The supply-side math makes sense if you freeze demand. But demand doesn't freeze. When something that was expensive becomes cheap, every application that was previously unaffordable becomes viable. The total need for the thing expands, often enormously.
Software has been expensive — in time, in expertise, in organizational prioritization — and that expense has been quietly rationing demand for decades. Cheap AI-assisted development doesn't eliminate that demand. It unlocks it.
// The 12 to 18 Month Gap
I want to be honest about the near term, because I think the near term is going to be genuinely rough and it is not useful to paper over it with optimism.
Right now, organizations are doing exactly what Cloudflare did. They're looking at support queues, at entry-level roles, at functions that were always human by default rather than by design, and they're running the math. The math says cut. And so they cut. This is the right-sizing moment, and it is happening broadly, and it is going to continue for some time.
Most organizations haven't yet figured out what the new demand looks like. They can see the cost reduction opportunity clearly; the new value creation opportunity is harder to see because it requires imagining things that don't exist yet. The gap between "AI can replace this function" and "here is the new thing we are building that this function enables" is not trivial to cross. It requires leadership that can hold both ideas at once.
In the next year to eighteen months, a lot of the headlines are going to sound like Cloudflare. That's real, and the people affected are real, and I am not going to pretend otherwise.
But I've also started to see the other story beginning to take shape.
// The Forward Deploy Engineer
There is a role that has been gaining traction in Silicon Valley over the last year, and I think it is a leading indicator of where this goes. The title varies — forward deploy engineer, customer success engineer, embedded engineer — but the pattern is consistent.
The traditional B2B SaaS model has a gap in it. You build a platform. Your customers buy access to the platform. What they actually need is software that fits their specific workflows, their data shapes, their operational context — but bespoke software was always an agency problem, a consulting problem, an expensive custom engagement that most customers couldn't justify and most vendors couldn't operationalize at scale.
The forward deploy engineer closes that gap. These are engineers embedded with customers, building custom software on top of the vendor's core product — in real time, in context, at a cost structure that AI-assisted development has made viable. The result is software that fits the customer precisely, which means the customer is more dependent on it, which means the vendor's offering becomes more entrenched and stickier than pure platform access ever made it.
This is a new category of engineering work. It didn't exist as a scalable practice when bespoke development required months and five-figure contracts. It works now because a forward deploy engineer with good AI tooling can produce in a day what used to take a consulting engagement weeks to scope. The unit economics changed. The role became viable.
Silicon Valley is still figuring out how to fully characterize this role — what it costs, how it's staffed, where it sits on an org chart. But the trajectory is clear, and the need it's filling is real.
// The Internal Software Backlog That Never Shipped
There is another enormous demand category that tends to get overlooked in conversations about AI and software development, because it's invisible. It's invisible because it has never been built.
Think about the internal tools that exist in any organization above a certain size. The dashboards that are actually maintained. The monitoring, the reporting, the automation for processes that are currently handled manually or by spreadsheet. The support tooling that was never scoped because no engineering sprint ever had room for it. The onboarding workflow that everyone agrees is broken but that hasn't made it onto a roadmap in four years.
This backlog is massive. It exists in every company. It never gets prioritized because internal tooling competes for engineering resources against customer-facing features, and customer-facing features always win. The internal stuff wasn't urgent enough, wasn't revenue-bearing enough, wasn't worth the cycle time of a proper sprint.
AI-assisted development changes the denominator dramatically. Internal tools that would have required a full sprint can now be produced in an afternoon. Dashboards that needed a dedicated frontend engineer can be generated from a description. Automation that required someone who knew how to write code now requires someone who knows enough to describe what they want clearly.
The bar to entry has dropped far enough that the internal software backlog is suddenly addressable — by smaller teams, by less specialized people, in smaller increments. This is another enormous pool of demand that was rationed by cost. The cost just changed.
// What Holds Through the Transition
I've written before about the conductor metaphor — the idea that the programmer's role is shifting from the person who writes the music to the person who leads the performance. What I want to add here is a practical implication that I think the Jevons framing makes clearer.
The organizations that survive the next eighteen months well are the ones that recognize the difference between the cost center and the knowledge carrier. Support staff aren't just people who close tickets. They're repositories of contextual intelligence about your customers, your product, and the gap between the two. That intelligence is the thing that trains the chatbot, informs the designer, and makes the forward deploy engineer effective. If you jettison it to hit a Q3 number, you've traded a long-term asset for a short-term saving.
IKEA understood that their support people had a rare and expensive thing, and that AI had given them an opportunity to redeploy it rather than a reason to discard it. That's not sentimentality. That's strategy.
The engineers who are watching this unfold and wondering where they fit have a version of the same opportunity. The skills that made you good at your job — knowing how systems connect, understanding what users actually need versus what they say they need, being able to hold a complex problem in your head and navigate it — those skills don't depreciate when AI handles the implementation. They become the scarce resource that makes the AI useful.
// Where This Goes
The short-term story is going to be painful and loud and real. Layoffs are real. Uncertainty is real. The people trying to figure out how to position themselves in a market that is actively reorganizing underneath them have a legitimately hard problem.
But the medium and long-term story — the Jevons story — is one of expansion, not contraction. Software is not going away. It is getting cheap enough to go everywhere it was previously too expensive to go. The demand surface is not shrinking. It is about to be larger than it has ever been.
The forward deploy engineer is one shape of that. Internal tooling finally getting built is another. New business lines that emerge when deep product knowledge gets redirected rather than discarded — IKEA's billion dollars — is a third.
We are in the gap between the thing that was and the thing that will be, and the gap is uncomfortable. I don't want to minimize that. But the gap is not the destination. And for anyone who wants to make a place for themselves in whatever the destination looks like, the window is open. The tools are available. The demand is forming, even where it isn't yet visible.
Jevons watched an efficiency gain and correctly predicted that it would produce more consumption, not less. He was describing coal in 1865. The principle holds.
The thing that got cheaper is about to be everywhere.