Because it’s about people, not procurement.
MIT researchers found that 95% of businesses are failing to see meaningful returns from AI. A separate study showed that experienced developers using AI tools actually took 19% longer on tasks than those who didn’t use them at all. Daron Acemoglu (a Nobel laureate) looked at the macro picture and landed on 0.5% productivity improvement over the next decade.
And yet, many practitioners are seeing gains that are not 20%, nor 2X, but orders of magnitude improvements. Small teams are shipping what used to require departments. Solo builders producing in a week what used to take a quarter.
Both things are true at the same time, so it’s important to understand how & why that dichotomy exists.
There are levels to this AI thing, and you need to be honest about where you are…
Picture the kids in Stranger Things riding around Hawkins on their bikes, high-fiving each other after every mission. To them, they’re moving fast…and compared to walking, they are. Meanwhile, somewhere above them in outer space, aliens are zipping around in spaceships, and they’re completely oblivious.
First, let’s look at the what the AI adoption landscape looks like
Let’s look deeper at each level:
Likewise, it’s important to understand how the tech changes at each level
L0 → L2: Ground transport (tools-based)
The MIT data is likely measuring teams at L0 & L1, because this is where virtually everyone is right now. Those studies aren’t wrong. They’re just clocking the kids in Hawkins and concluding that personal transportation has modest upside.
L0 through L2 are largely procurement, and training, decisions. Your IT department can get you there by buying tools, provisioning access, hosting training, measuring utilization, and then they declare victory. It’s the same playbook as every enterprise software rollout in history. Most organizations that think they’ve “done AI” when they buy the right tool, and mistake motion for momentum.
That said, you can get big gains here, particularly when you get to L2. This is where you see engineers ditching IDEs entirely and able to ship full products without ever opening one. Honestly, most orgs aren’t even here yet.
L3: Getting initially airborne
The discontinuity in the curve starts as you move from L2 to L3. This is the first step that can’t be procured. It’s the moment you cross from “we bought a thing” to “we changed how we work.”
This is where Storm Thinking kicks in. Buying more tools feels like progress. It’s auditable, it’s defensible, it shows up green on a QBR slide. However, actually changing how a team works is harder to measure and scarier to own…especially when you have a team of 100+ and there are no established playbooks.
This is where all stages of the product development process must adjust how they work. Designers are no longer building static screens in Figma, but fully coded prototypes. PMs stop writing traditional PRDs, and start developing tight, machine readable specs an autonomous agent can work from. Engineers stop thinking of themselves as builders who use AI, and start thinking of themselves as systems designers who review AI output.
The Plan → Build → Test → Verify → Retro loop is a major process change; however, still at the scope that an individual engineering leader can implement within their span of control. They need little to no executive buy-in, no cross-functional steering committees, no IT procurement cycles. Just a team willing to work differently and a leader with the conviction to change how they operate.
I’d put myself at L3; I’m “airborne” now, and understand what it means when the process starts orchestrating instead of the team. The hard part now, is how to get to L4, as that’s where it becomes genuinely challenging to enact at-scale. The reason is that it’s not about you anymore.
L4: Changing the system
Going from a helicopter to getting a commercial pilot’s license isn’t something a single engineering leader can pull off alone. It now requires the entire product development org to transform its staffing, processes, and tools.
None of that happens inside one team. It requires a CTO / VP to be hands-on, and change how the whole org (potentially thousands) builds, and then support it with instituting new processes, new tooling decisions, and the willingness to rethink how they structure, staff, and measure their teams.
At this level the challenges are organizational, not technical: how you govern agent access to production systems, how you restructure the SDLC across functions that have never had to coordinate this tightly, and how you redefine roles when the line between PM, designer, and engineer starts to blur. The tooling is the easy part. Getting design, product, and engineering to fundamentally change what they produce (and how they hand it off to each other) is the actual work.
L5: The conversation no one is having
Becoming a team of astronauts, piloting a fleet of spaceships, is likely something few CEOs are talking about, as it forces questions most organizations aren’t ready for.
For instance, if one engineer running agentic workflows can produce what ten engineers used to, what does your headcount model look like? What skills do you recruit for? What happens when one person at the same level performs 10x their peer Do you promote them even after 6 months at level? Do your comp bands even account for that? What does your capex/opex split look like when your most leveraged asset is a $20/month API subscription and the human reviewing its output?
These aren’t hypothetical problems. They’re arriving now. And most CEOs are still thinking about AI as a productivity multiplier on top of their existing org structure, rather than as something that calls the org structure itself into question.
We’ve already seen examples of engineers working multiple jobs (very well), because the current process doesn’t adequately incentivize them. That’s neither just a loyalty, ethics, or legal issue, but a market signal that the current incentive structure is obsolete.
This is The Disruptor’s Paradox playing out in real time. The organizations with the most resources to make this transition are also the most invested in preserving the structure that got them here. So they optimize. They buy tools. They measure utilization. And they slowly, confidently, fall behind the teams small enough to just decide to work differently.
It’s “February 2020” in the tech industry…remember those days??
Smaller orgs have an “unfair” advantage
Smaller teams and startups often don’t feel this friction at every step, as can move faster. It’s not that the tools work differently, it’s that one person can span multiple levels of decision-making, and there are fewer existing processes, and investments, to undo.
The founder, who is also the CTO and engineering manager, can move from L1 to L4 in an afternoon. No alignment tax. No steering committee.
In a large org, those are four different people in four different conversations that have to happen in sequence. That’s where the drag lives. And that’s why the productivity gains show up so clearly at the edges of the industry, such as indie developers, startups, and small teams…meanwhile enterprises need to overcome that inertia.
What this means if you’re trying to push things forward
If you’re an IT leader: You’ve done your part when you’ve guided L0 through L2. If you’ve delivered that, it’s a real gain, and you’re probably ahead of 95% of companies already. But understand the ceiling. The next step isn’t more tools. It’s handing the baton to engineering leadership and making sure you empower them to run with it by clearing security, compliance, and operational barriers.
If you’re an engineering manager: L3 is your call. You don’t need permission. You need a team willing to work differently and the conviction to change how you operate. The path is visible: agentic tools, structured loops, getting engineers comfortable stepping back from every keystroke. Go do it, and be successful with it.
If you’re a CTO / VP: L4 is yours. It requires you to coordinate all of the product teams and all of the functions to fundamentally change the way they work, as well as to adjust the processes, tools, and team structures to make it happen. Skip the pilots, the working groups, and make the decisions now.
If you’re a CEO: Nobody is likely coming to you with the L5 conversation (yet)…but it’s coming. The organizations that get to L4 will need dramatically different team structures, hiring profiles, and financial models than the ones running today. That’s no longer a technology decision, but a company design decision. The CEO and Board are the only ones who can make that happen.
Aliens and spaceships exist, but they live on the other side of a transition that most companies aren’t even aware of yet, far less are laying the groundwork for that shift.
Figure out where you are today, who owns the next level at your company, and then empower them to act.