Most Companies Are Running AI Blind. A New Report Says Context Is the Fix.
Only 16% of organisations are getting transformative value from AI, according to a poll published by MIT Technology Review Insights. The problem isn't the models. It's that the models know almost nothing about the business they're working inside.
Key points
- Only 16% of organisations surveyed by MIT Technology Review Insights report transformative value from their AI investments, as of 2026.
- 23% of those same organisations say they are seeing minimal returns or none at all.
- Just 12% of enterprises have AI agents, software that can carry out multi-step tasks on its own, running in real production environments.
- AI experts describe the missing piece as a "context layer": an architectural bridge that tells AI systems what a specific business actually does and how it operates.
- AI2Day first covered the context-layer problem on 14 July 2026 and has returned to it twice since.
Picture the smartest temp worker you have ever hired. Now blindfold them and push them through the front door of your office. That, roughly, is what companies are doing when they deploy AI agents without giving those agents any real knowledge of the business.
A poll conducted by MIT Technology Review Insights captures the gap in hard numbers. Every organisation surveyed was already using AI in some form. Yet only 16% reported transformative results. Forty percent said they were seeing moderate gains in specific tasks, and 23% reported minimal returns or none at all.
Why aren't the results better?
Powerful models aren't the bottleneck. Most businesses plug AI into their operations without grounding it in how that specific company actually works, and the results reflect that gap.
Prukalpa Sankar, co-founder of data-management platform Atlan, puts it plainly in the report. "There's a big gap between what the models are capable of and how the models are actually being useful inside organizations," she says. "For every real-life intent and purpose, artificial general intelligence is already here. It's just not really useful in most companies today."
What's missing, Sankar argues, is a context layer: architectural infrastructure that sits between a company's internal systems and whatever AI tools that company is running. It translates internal knowledge into something AI agents can actually act on. Cut that bridge and an agent can be impressively capable in general while being useless in practice, because it doesn't know what your product codes mean or who approves purchase orders.
"Without a context layer for AI, it's almost like you're hiring the smartest intern in the world, blindfolding them, and throwing them into your organization," Sankar says.
What are companies actually doing with AI right now?
Mostly still experimenting. Thirty-eight percent of respondents say they are running pilots, and 35% are using AI tools like co-pilots or chat assistants to help individual employees. Just 3% say AI is doing most of the work in any key process.
| Stage of AI use | Share of respondents |
|---|---|
| Piloting use cases | 38% |
| Employee co-pilots / chat assistants | 35% |
| Meaningful gains | 21% |
| Agents in production | 12% |
| AI doing most work in key processes | 3% |
That 12% figure for agents in production is telling. We've tracked the surge in enterprise AI coverage closely: our story "Enterprise AI took over our week, and the shape of it says something" found that AI2Day published 11 enterprise AI stories between 21 and 27 October 2026, up 175% from the week before. One theme kept appearing across those pieces: agents that work well in demos fall short when released into a real, messy enterprise environment.
The context problem explains a lot of that. An agent helping a bank process a loan application needs to know that bank's specific rules and data definitions. A general-purpose model doesn't carry that knowledge by default.
The honest read here is that most organisations have bought into AI faster than they've built the foundations that would make it work. Vendors sell capability; nobody sells the unglamorous infrastructure work of teaching a model what your company actually is. Until that changes, the gap between what AI can do and what AI is doing will stay wide.
Common questions
Does this mean AI isn't ready for business use?
No. It means businesses aren't yet ready for AI. The models are capable; the internal plumbing, specifically the systems that feed AI agents accurate, company-specific knowledge, usually isn't built yet.
What should an employee or manager do if their company's AI tools seem unhelpful?
Ask whether the tool has been given access to your company's actual data and processes. Generic AI tools applied to specific business problems without that grounding will almost always disappoint.
