The Efficiency Trap

Wrapping agents around old workflows fossilises an operating model built for human constraints. Invest in the human capabilities that compound instead.

The Efficiency Trap is what organisations fall into when they default to using AI to do the same work as before, but better, faster and cheaper.

And when those AI initiatives give people time back, they spend it doing more of that same work. The result is an obsolete operating model running at speed.

Efficiency itself isn’t the problem. The problem is when efficiency becomes the only strategy, crowding out the exploration that creates new value.

This article picks up one specific failure mode inside that trap. Two speakers at this year’s SXSW addressed it from different angles. First, Ian Beacraft said that “…most organisations are optimising for a world that no longer exists”. Ooof.

Second, Brian Solis said that “…most organisations are using AI to fossilise yesterday’s business model”. He went further and put a figure on the cost of this error: 40% of AI productivity gains are being lost to what he calls the ‘AI tax’, the friction of bolting agentic capability onto workflows it was never built to fit. The correcting, the checking, the rework.

This misstep is the fossilisation move.

What fossilisation actually does

Previous waves of digital tech sat on top of the business. You could rip out a CRM, swap a website or replace a tool, and the underlying way work happened stayed broadly intact, because the structures of work weren’t going anywhere.

Ian Beacraft talks about organisational physics. For 150 years, organisations have been shaped by three durable human constraints:

Expertise is scarce.
Attention is finite.
Knowledge takes time to scale.

Almost every structural feature of corporate life is a response to one of those.

Approvals exist to manage risk and distribute accountability. Departments exist because expertise needs a community to live in. We have sequential workflows because human attention is, well… what human attention is…

Ooo, shiny!

And front-loaded onboarding exists because teaching in context isn’t feasible at scale.

None of that is incidental. It evolved from how risk gets managed, how authority gets distributed and how workload gets routed.

Now wrap agents around the workflows that result. You’re encoding decision logic, escalation patterns and tacit context into a system that quietly becomes the new infrastructure. But you’re encoding it on top of structures that were built to manage human constraints the agentic layer doesn’t have.

Agents don’t run out of attention. They don’t need expertise to live in a community. They don’t need knowledge to scale slowly.

So work works differently.

The workaround becomes the architecture, just as the thing it was working around disappears.

Pull the system out a year later and you lose more than the agents. You lose a year of organisational learning baked into them.

That’s why fossilisation is the right word. You’re preserving the current operating shape of the business in a material that’s much harder to reshape than anything before, at the precise moment that shape has the most reason to change.

What gets quietly evacuated

There’s a second cost, and it’s less visible than the first.

Sam Jordan’s argument was that manual processes carry human value as a by-product.

Intuition, curiosity, the read on a room, institutional memory, character: all of this perspective, context and skill accumulates as a by-product of doing the work the ‘slow’ way.

None of it sits in a workflow diagram. All of it sits within the people doing the work.

Klarna is the case study everyone usually reaches for (replacing a large share of its customer service agents with automation). The narrative frames it as an AI failure. I don’t think it was.

I imagine the technology worked fine. But I also imagine that, in a rabid thirst for automation, nobody paused to ask what the emotional shape of those customer service interactions was, and therefore what would be lost when the organisation shot for automation and fell into the Efficiency Trap.

The agents replaced the workflow, but they didn’t replace the by-product, so a chunk of customer relationship efficacy and EQ left the building.

Replace the workflow without understanding the by-product and designing it back in, and you’ll fall into the Efficiency Trap.

Where AI investment is actually landing

Pause on what AI is good at, and what it isn’t.

AI excels at executing well-defined tasks against rich data: scale, speed and pattern matching.

It’s much weaker at the things that don’t show up in the dataset: context, motivation, the why beneath the what.

When designing experiences at ELSE, we use a six-layer ladder to frame the link between the what and the why. I think it’s a useful framework for AI initiatives aimed at internal work, too.

At the top sits context, then motivation, then goal. These are strategic, durable, emotional and almost impossible to copy.

For example, in a consumer context, a customer of financial services from a comfortable household will hold a motivation towards financial security, with underlying goals around becoming mortgage-free and building a rainy-day fund. Maybe they’re trying to live modern life by the 50/30/20 rule (50% on needs, 30% on wants, 20% put aside).

Our goals may shift over time, but motivation rarely does.

At the bottom of the stack sit task (the thing I do in pursuit of the goal), features (the enablers of the task) and interactions (the up-close doing). These are executional, functional and fully commoditisable.

The line between those two halves matters for AI product strategy. A lot of agentic AI investment is landing almost entirely on the bottom half, automating tasks, features and interactions, built on assumptions about context that get misread.

This is where the fossil forms.

The augmentation move

Solis frames the alternative as ‘Innovative AI’ rather than ‘Iterative AI’. At ELSE we’ve called it AI-retrofit vs AI-native, but we could simply call it augmentation rather than automation.

Automation puts agents to work delivering existing workflows.

Augmentation starts from the human capabilities you want to elevate, then designs the agentic layer to make those capabilities more possible.

The capabilities worth designing for are the ones that compound with experience and don’t commoditise.

I like Kevin McCullagh’s framing of vision, judgement and craft:

Vision: seeing where the business needs to go before the evidence makes it obvious.
Judgement: the call on what’s worth doing, in what order, for whom.
Craft: the tacit knowledge of making something well.

To those I’d add two more. They’d probably sit under judgement in Kevin’s framing, but I want to call them out:

EQ: the ability to read a room and sense the emotional reality under a stated position.
Intuition: pattern recognition at speed, earned through experience.

These five live at the top of the ladder, and they map to context, motivation and goal. An AI strategy that doesn’t make them more possible is investing at the wrong end.

Three moves

Reopen your understanding of context. Almost everyone’s operating context has shifted. Few teams have redesigned from the changed position. Agents that help you see the new context are worth more than agents that execute against the old one.

Spend freed capacity at the top. When AI gives people time back, the fossilising instinct is to do more at the task layer. The augmenting, AI-native instinct is to push deeper into the why.

Name the human acts. Be explicit about which capabilities your product exists to elevate, and design the agentic layer to make those capabilities more possible. Don’t focus on task execution speed and accuracy alone. If you can’t name the elevations you’re designing for, you don’t have a strategy. You have a feature list.

The question to leave with

One question for any leader currently shipping agentic capability:

You’re investing in AI to go faster, but are you also investing in AI to go somewhere new?

Both are legitimate moves, but I’d wager only the second will still be worth doing in three years.

The fossil record is full of species that were brilliantly adapted to a world that was disappearing around them.

Next in the series: Leadership Will Break in the Agentic Era.