# Yoav Shoham argues in MIT Technology Review that AI-agent hype risks backlash [Canonical reading](https://escapemeridian.com/readings/yoav-shoham-argues-in-mit-technology-review-that-ai-agent-hype-d0848389) On July 3, 2025, Yoav Shoham argues in MIT Technology Review that excitement about AI agents could outrun their ability to work reliably, inviting disappointment and backlash. He says “agent” is applied to systems ranging from simple scripts to sophisticated AI workflows, leaving unclear what they can do, how autonomously they operate, and how reliably they perform. He also warns that large language models (LLMs) can make things up or fail during multistep tasks, and recounts an automated Cursor support agent inventing a device-use policy that did not exist. Shoham argues that reliable agents need systems that account for uncertainty, monitor outputs, manage costs, and add guardrails. Cooperation presents another problem: Google’s A2A protocol describes how agents communicate, he says, but not what their words mean; agents representing competing interests may also lack reason to cooperate. His conclusion is not that the idea should be abandoned: he says these problems can be addressed, but calls for thoughtful design, clear definitions, and realistic expectations. Otherwise, he warns, users and developers may discover the pitfalls the hard way and develop a negative view of the whole endeavor. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source MIT Technology Review · Article Original source date: Jul 3, 2025 (2025-07-03T09:00:00Z). Date basis: Date from the publisher’s own records, through its API. Added to Escape Meridian: 2026-11-13T06:03:24.430Z. First observed: 2026-10-31T18:30:12.184Z. Retrieved: 2026-10-31T18:31:57.164Z. - [Don’t let hype about AI agents get ahead of reality](https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/) ## Limits Text as retrieved on October 31, 2026; originally published July 3, 2025. ## Open question How can agents from different vendors share enough meaning and align incentives to coordinate reliably? ## Source-stated Yoav Shoham argues that hype about AI agents could get ahead of their ability to deliver real benefits, risking backlash. > if we’re not careful, we’re going to derail the whole idea before it has a chance to deliver real benefits. As with many tech trends, there’s a risk of hype racing ahead of reality. - [Supporting source](https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/) ## Source-stated Shoham says there is no shared definition of “agent,” which can leave customers confused about what systems can do and how reliably they perform. > There’s no shared definition, which leaves plenty of room for companies to market basic automation as something much more advanced. - [Supporting source](https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/) ## Source-stated Shoham says large language models can make things up or go off track, especially on multistep tasks using external tools and chained responses. > They can make things up, go off track, or fail in subtle ways—especially when they’re asked to complete multistep tasks, pulling in external tools and chaining LLM responses together. - [Supporting source](https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/) ## Source-stated Shoham argues that systems built around LLMs should account for uncertainty, monitor outputs, manage costs, and include safety and accuracy guardrails. > systems that account for uncertainty, monitor outputs, manage costs, and layer in guardrails for safety and accuracy. - [Supporting source](https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/) ## Source-stated Shoham says Google’s A2A protocol defines how agents communicate but not what they mean, making coordination brittle without shared vocabulary or context. > It defines how agents talk to each other, but not what they actually mean. - [Supporting source](https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/) ## Source-stated Shoham warns that agents representing different vendors, customers, or competitors may not have aligned incentives to cooperate. > Without some way to align incentives through contracts, payments, or game-theoretic mechanisms, expecting seamless collaboration may be wishful thinking. - [Supporting source](https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/) ## Source-stated Shoham concludes that the potential of agents is real, but says thoughtful design, clear definitions, and realistic expectations are needed to avoid a negative view of the endeavor. > The potential here is real. But we need to match the ambition with thoughtful design, clear definitions, and realistic expectations. - [Supporting source](https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/) ## Source and contributor context MIT Technology Review · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## AI agent An AI system that chooses and carries out actions toward a goal, often using software tools. - [Definition reference](https://www.ibm.com/think/topics/ai-agents) ## Large language model A language model with many learned numerical settings, used to process and generate language. - [Definition reference](https://developers.google.com/machine-learning/glossary#large-language-model) ## Cursor Cursor makes an AI coding environment and publishes evaluations of agents working on software tasks. - [Explore this context](/entities/cursor) - [Reference](https://cursor.com/) - [Reference](https://cursor.com/evals) - [Reference](https://cursor.com/blog/real-time-rl-for-composer) ## Editorial provenance Version 1. Publication ID: 0fbfdf41-a59a-6690-b1dc-db890e6451e7. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-13T05:59:25.712Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical. Observation: 45620c135ab7fe8d28b11f699f5f1bd4b9cd0c0e015866f4d1b8157f68e18636. Source text hash: 95d25138eac4297522abaa64166f02b71d250bb8329e2de715148406b2f961f7. ## Revision and source-check history ### Version 1 — 2026-11-13T06:03:24.430Z On July 3, 2025, Yoav Shoham argues in MIT Technology Review that excitement about AI agents could outrun their ability to work reliably, inviting disappointment and backlash. He says “agent” is applied to systems ranging from simple scripts to sophisticated AI workflows, leaving unclear what they can do, how autonomously they operate, and how reliably they perform. He also warns that large language models (LLMs) can make things up or fail during multistep tasks, and recounts an automated Cursor support agent inventing a device-use policy that did not exist. Shoham argues that reliable agents need systems that account for uncertainty, monitor outputs, manage costs, and add guardrails. Cooperation presents another problem: Google’s A2A protocol describes how agents communicate, he says, but not what their words mean; agents representing competing interests may also lack reason to cooperate. His conclusion is not that the idea should be abandoned: he says these problems can be addressed, but calls for thoughtful design, clear definitions, and realistic expectations. Otherwise, he warns, users and developers may discover the pitfalls the hard way and develop a negative view of the whole endeavor. ## Original source MIT Technology Review · Article Original source date: Jul 3, 2025 (2025-07-03T09:00:00Z). Date basis: Date from the publisher’s own records, through its API. Added to Escape Meridian: 2026-11-13T06:03:24.430Z. First observed: 2026-10-31T18:30:12.184Z. Retrieved: 2026-10-31T18:31:57.164Z. - [Don’t let hype about AI agents get ahead of reality](https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/) ## Limits Text as retrieved on October 31, 2026; originally published July 3, 2025. ## Open question How can agents from different vendors share enough meaning and align incentives to coordinate reliably? ## Source-stated Yoav Shoham argues that hype about AI agents could get ahead of their ability to deliver real benefits, risking backlash. > if we’re not careful, we’re going to derail the whole idea before it has a chance to deliver real benefits. As with many tech trends, there’s a risk of hype racing ahead of reality. - [Supporting source](https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/) ## Source-stated Shoham says there is no shared definition of “agent,” which can leave customers confused about what systems can do and how reliably they perform. > There’s no shared definition, which leaves plenty of room for companies to market basic automation as something much more advanced. - [Supporting source](https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/) ## Source-stated Shoham says large language models can make things up or go off track, especially on multistep tasks using external tools and chained responses. > They can make things up, go off track, or fail in subtle ways—especially when they’re asked to complete multistep tasks, pulling in external tools and chaining LLM responses together. - [Supporting source](https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/) ## Source-stated Shoham argues that systems built around LLMs should account for uncertainty, monitor outputs, manage costs, and include safety and accuracy guardrails. > systems that account for uncertainty, monitor outputs, manage costs, and layer in guardrails for safety and accuracy. - [Supporting source](https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/) ## Source-stated Shoham says Google’s A2A protocol defines how agents communicate but not what they mean, making coordination brittle without shared vocabulary or context. > It defines how agents talk to each other, but not what they actually mean. - [Supporting source](https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/) ## Source-stated Shoham warns that agents representing different vendors, customers, or competitors may not have aligned incentives to cooperate. > Without some way to align incentives through contracts, payments, or game-theoretic mechanisms, expecting seamless collaboration may be wishful thinking. - [Supporting source](https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/) ## Source-stated Shoham concludes that the potential of agents is real, but says thoughtful design, clear definitions, and realistic expectations are needed to avoid a negative view of the endeavor. > The potential here is real. But we need to match the ambition with thoughtful design, clear definitions, and realistic expectations. - [Supporting source](https://www.technologyreview.com/2025/07/03/1119545/dont-let-hype-about-ai-agents-get-ahead-of-reality/) ## Source and contributor context MIT Technology Review · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## AI agent An AI system that chooses and carries out actions toward a goal, often using software tools. - [Definition reference](https://www.ibm.com/think/topics/ai-agents) ## Large language model A language model with many learned numerical settings, used to process and generate language. - [Definition reference](https://developers.google.com/machine-learning/glossary#large-language-model) ## Cursor Cursor makes an AI coding environment and publishes evaluations of agents working on software tasks. - [Explore this context](/entities/cursor) - [Reference](https://cursor.com/) - [Reference](https://cursor.com/evals) - [Reference](https://cursor.com/blog/real-time-rl-for-composer) ## Editorial provenance Version 1. Publication ID: 0fbfdf41-a59a-6690-b1dc-db890e6451e7. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-13T05:59:25.712Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model, second\_opinion\_categorical. Observation: 45620c135ab7fe8d28b11f699f5f1bd4b9cd0c0e015866f4d1b8157f68e18636. Source text hash: 95d25138eac4297522abaa64166f02b71d250bb8329e2de715148406b2f961f7. ### passed — 2026-11-13T05:59:25.712Z — anthropic/claude-opus-5.5 Publication: 0fbfdf41-a59a-6690-b1dc-db890e6451e7 attribution — pass: The reading consistently attributes the argument to Yoav Shoham writing in MIT Technology Review, matching the byline. Arguments and the Cursor example remain his claims. It does not mention his AI21 affiliation or the Maestro plug, but the reading does not rely on that passage, so attribution is not distorted. > Yoav Shoham is a professor emeritus at Stanford University and cofounder of AI21 Labs. summary — pass: The summary leads with the actor and thesis: hype risks a backlash. It covers the source's three problems: definitions, LLM reliability with the Cursor example, and A2A semantics and incentives. It keeps his non-abandonment conclusion and his warning. AI agent and LLM are defined on the page. A2A is described in plain terms. The question follows directly from his coordination argument. > None of these issues are insurmountable. Shared semantics can be developed. Protocols can evolve. date — pass: July 3, 2025 matches the UTC day of originalPublished. The limitation notes the later retrieval date. The page was modified on 2025-07-15, but the cited passages are plainly part of the op-ed body. Related-story blurbs later on the page describe later events, such as AlphaGo plus ten years. The reading neither uses nor relies on them. No hindsight framing appears, and forecasts stay attributed. > Ten years after AlphaGo’s match against Go champion Lee Sedol, today’s AI still isn’t tapping into the machinery support — pass: Each claim's excerpt is an exact source passage that supports it. The edit flag does not affect them, because they are core body text of the original op-ed. The incentive and cooperation points are attributed as Shoham's warnings. No benefit, cost or forecast is presented as fact. > There were widespread complaints and reports of users canceling their subscriptions. But it turned out the policy didn’t exist. The AI had invented it. media — pass: No media is selected, and no verified media candidates were supplied. That is appropriate. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/d084838965e76ee47e63a7faf0380b826b1047833872869e2ffec7525083fed5/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/yoav-shoham-argues-in-mit-technology-review-that-ai-agent-hype-d0848389.json)