MIT researchers have put a number on something business leaders have been arguing about for years: how much of the workforce can AI actually replace right now?

Their answer, published in a new study this month, is 11.7% of the US labor market. In wage terms, that is approximately $1.2 trillion in annual compensation that AI systems are already capable of handling at an acceptable standard.

The researchers called their framework the Iceberg Index — a name chosen deliberately. Like the bulk of an iceberg sitting below the waterline, most of AI’s labor market impact is not yet visible in unemployment data or hiring freezes. It is accumulating beneath the surface while organisations figure out how and whether to deploy it.

What the Index Actually Measures

The Iceberg Index does not ask “will AI take jobs?” It asks a more practical question: across specific workplace tasks, can AI perform them at a minimally acceptable standard today?

The answer varies significantly by sector.

Legal work is the hardest for AI to replace. Researchers found AI achieves only a 47% success rate on legal tasks — too low for serious deployment in most firms. The need for precise judgment, strategic guidance, and contextual interpretation makes legal reasoning one of the tougher problems for current models.

Installation, maintenance and repair work showed the highest AI success rate at 73%. This sounds counterintuitive — physical jobs are often thought to be safer from automation — but the study is measuring AI’s ability to handle the documentation, diagnostic, and planning components of these roles, not the physical execution itself.

Media, arts and design sit in the middle at around 55%. AI proves useful for drafting and iteration but falls short on high-end creative execution.

A Rising Tide, Not a Crashing Wave

The researchers used a phrase that should shape how business leaders think about this: AI’s workforce impact is advancing like a rising tide, not a crashing wave.

A crashing wave hits specific sectors hard and suddenly — think of how streaming wiped out video rental overnight. A rising tide moves gradually across everything. It changes the composition and character of work broadly, rather than eliminating specific job categories in one move.

When AI becomes capable of performing most tasks within a given role, employment in that role tends to fall by about 14%, the research found. But when AI’s capability is concentrated in just a few tasks within a job, employment in that role can actually grow — because workers become more productive overall, and companies hire more of them to meet demand.

This is the nuance that gets lost in most AI-and-jobs coverage. The effect is not binary.

The Trajectory Ahead

The study also modelled the pace of change. In 2024, AI models could complete roughly 50% of text-based tasks at a minimally acceptable level. By 2025, that figure had risen to around 65%. At the current rate of capability improvement, AI could handle 80 to 95% of text-based tasks by 2029.

That does not mean 80-95% of knowledge workers lose their jobs. It means the ceiling on AI assistance across most desk work rises dramatically within this decade. The question for organisations is not whether to integrate AI, but how quickly and across which functions to do so with intention.

What This Means for Business

The Iceberg Index research reframes the AI-and-jobs conversation in a useful direction for business operators.

The threat is not mass unemployment — yet. The 11.7% figure represents AI capability today. Translating capability into actual job displacement requires companies to make deliberate decisions to restructure, retrain or reduce headcount. Most are still in early deployment stages. The data from other 2026 surveys confirms this: only about 10% of organisations have actually scaled AI agents across their operations.

The real risk is competitive lag. If 11.7% of the work in your business can already be done by AI at an acceptable standard, the question is whether your competitors are deploying that capability while you are still running manual processes. The PwC 2026 AI Performance Study found that 74% of AI’s economic gains are being captured by just 20% of organisations. The gap between leaders and laggards is widening, not closing. A recent enterprise survey by WRITER put the numbers on this more starkly: only 29% of organisations report significant ROI from AI despite widespread tool adoption.

Role composition matters more than headcount. The finding that AI concentrated in a few tasks can actually increase employment is important. Organisations that use AI to eliminate the most repetitive and low-value work within roles — and then redirect workers toward higher-judgment activity — often end up hiring more, not fewer. The firms most at risk are those running parallel staffing: paying humans to do tasks AI could handle, without capturing the productivity upside in either workforce costs or output quality.

Skills investment is a strategic hedge. The legal sector’s resistance to AI replacement (47% success rate) comes from the complexity and judgment required — skills that take years to develop. Organisations that invest in building genuine analytical and strategic capability in their teams are building the part of the workforce that is hardest to automate. Data literacy, AI fluency, and complex problem-solving are not just nice-to-haves — they are structural defences against the rising tide. The research on data-literate organisations shows they are categorically better at leveraging AI tools precisely because their teams can evaluate and direct AI outputs rather than accept them blindly.

The Practical Move

The Iceberg Index is a useful tool for any business leader trying to cut through the noise on AI and jobs. Stop asking “will AI replace my team?” and start asking: “Which specific tasks in my operation fall in the 47-73% AI capability range, and what does that mean for how I structure work, hire, and invest in training over the next 18 months?”

That is a tractable question. It leads to real answers and real decisions.

If you want to map your organisation’s AI exposure — understanding where you’re most at risk from competitors who are automating faster, and where your team’s skills create durable advantages — that is exactly the kind of strategic work the Omni Advisory team works through with clients.

For teams that want to build the analytical and AI skills that sit above the automation line, the Enterprise DNA learning platform covers data, AI, and business intelligence across every skill level.

The tide is rising. The question is whether your organisation is already on higher ground.