MIT Tech Review's Hype Index points at unsexy AI

The July 29 edition of MIT Technology Review's Hype Index opens with two stories that are hard to fit into the same sentence. On one side, an open letter signed by leading economists warning that AI may end up taking your job. On the other, the company 1X showing off a pair of remarkably dexterous humanoid robots along with the promise that they will soon be better than you at making dinner. The piece sums it up without mercy: insult to injury.

The headline of this edition, however, is not about humanoids. It is about the opposite: The AI Hype Index: Unsexy AI, the AI that does not photograph well.

What the Hype Index measures

It is a recurring MIT Technology Review section that sorts the AI news of recent weeks along two variables: how much noise each story generated and how much real substance sits behind it. It is not a product ranking or a buying guide, it is a thermometer for expectations. And expectations tend to be the worst managed variable in the AI projects that reach an agency.

The value of reading it lies in the contrast. When a robotics demo occupies the same mental space as a labour market report, the index forces you to separate two questions that usually travel together: what has been demonstrated in a controlled setting, and what already works in production with real users and real costs. A robot folding laundry in a two minute video and a system that classifies invoices every morning at six do not belong to the same category of evidence, even if they share a front page. That separation exercise is more useful than any five year prediction, because you can run it with the information already on your desk.

Why boring is the right word

The thesis of this edition is that the work actually changing how companies operate is not photogenic. Data extraction from documents, ticket classification, contract review, reconciling records that somebody used to handle by hand. None of it makes headlines, and almost all of it can be measured in hours saved the week after deployment.

The pattern has held for a couple of years now: the expensive part of an AI project is not the model, it is everything around it. Permissions, input formats, error handling, and a record of what happened and why.

The same thing happens in the Claude ecosystem. Most of what we have delivered this year would not carry a launch video:

1. MCP servers connecting a CRM or an ERP to Claude Code without exposing credentials,
2. PreToolUse hooks that block writes outside the project directory,
3. skills that package a client's tone and rules so nobody has to repeat them in every prompt,
4. subagents that review a report before anyone outside the team sees it.

None of those pieces is spectacular on its own. Together they are the difference between a proof of concept and something a team uses daily without thinking about it.

Who this reading is useful for

Anyone who has to decide a budget. If someone is weighing where to put the next quarter, the index works as a bias checklist: how much of what is under consideration comes from a demo, and how much from a verifiable deployment. It also helps technical profiles negotiating with management, because it names an intuition many already have and struggle to defend in a meeting.

One nuance is worth adding. The fact that 1X robotics is more demo than product today does not mean it will never arrive, just as the economists' letter is not a prediction but a warning about the speed of change. The index measures hype, not direction.

Our reading at ElephantPink is simple: the adoption clock is not set by the flashiest demo, it is set by the integration nobody shows in a keynote. If you have to choose where to spend effort this quarter, the answer is rarely on the cover.