# Ben Thompson argues in Stratechery that agents weaken the AI-bubble case [Canonical reading](https://escapemeridian.com/readings/ben-thompson-argues-in-stratechery-that-agents-weaken-the-ai-b6fddffb) Ben Thompson’s article was published on March 16, 2026. He argues that AI agents change the case for treating the industry’s spending as a bubble: agents can direct models, use tools to check results, and retry without a person managing each step. For coding and other verifiable tasks, he says, this addresses important weaknesses of chatbots and may make AI more useful. In Thompson’s account, each shift—from ChatGPT, to reasoning models, to agents—also increases compute use, while agents let a small number of people direct multiple systems. He argues that companies may use agents not only to reduce costs but to increase output, even as replacing workers with agents could make the transition “ugly.” Thompson also argues that an agent’s harness—the software that controls a model—can distinguish products, making integration between model and harness important to where profits flow. He sees this as a reason Anthropic and OpenAI may be more durable than critics of their valuations expect. His conclusion is firm: he thinks the investment is warranted and “we are not in a bubble.” Whether the expected demand, productivity gains and profits will materialize remains unknown. 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 Ben Thompson · Article Original source date: Mar 16, 2026 (2026-03-16T10:00:00+00:00). Date basis: Date stated on the publisher’s page. Added to Escape Meridian: 2026-11-11T02:23:40.890Z. First observed: 2026-11-10T14:13:44.551Z. Retrieved: 2026-11-10T14:15:35.703Z. - [Agents Over Bubbles](https://stratechery.com/2026/agents-over-bubbles/) ## Limits Text as retrieved on November 10, 2026; originally published March 16, 2026. Thompson’s discussion of agent capabilities, compute demand, productivity, layoffs, valuations and future profits is his argument and analysis, not evidence here that those outcomes have occurred. The article gives no measured basis for its claims about agents’ productivity effects or the resulting economic returns. It describes those effects as expectations and arguments. ## Open question What evidence will show whether agents deliver enough added productivity and economic value to sustain the compute investment Thompson describes? ## Source-stated Ben Thompson argues that agents can use tools to verify model output, reducing the need for users to manage mistakes in verifiable tasks such as coding. > the agent can also use other deterministic tools as well, which means that it can verify its results. - [Supporting source](https://stratechery.com/2026/agents-over-bubbles/) ## Source-stated Thompson argues that agents can increase compute demand while allowing a small number of people to direct multiple agents. > it actually won’t require that many people with agency to drastically increase the amount of compute that is actively utilized - [Supporting source](https://stratechery.com/2026/agents-over-bubbles/) ## Source-stated Thompson argues that agents could increase company output while replacing some workers, and says the consequences could be ugly. > There is a good chance this is going to get ugly; I’m not advocating for this outcome, rather analyzing why it is probably going to happen. - [Supporting source](https://stratechery.com/2026/agents-over-bubbles/) ## Source-stated Thompson argues that integration between a model and its harness can differentiate agent products and affect where profits flow. > the integration between model and harness is where true agent differentiation is found. - [Supporting source](https://stratechery.com/2026/agents-over-bubbles/) ## Source-stated Thompson concludes that the rise of agents makes him think the AI industry is not in a bubble and that the spending is warranted. > And, well, I think the rise of agents means we are not in a bubble. The capex is warranted, and Anthropic and OpenAI look more durable than ever. - [Supporting source](https://stratechery.com/2026/agents-over-bubbles/) ## Source and contributor context Ben Thompson · 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) ## Reasoning model A language model trained to work through a problem in written steps before it answers. - [Definition reference](https://en.wikipedia.org/wiki/Reasoning_model) ## Anthropic Anthropic develops Claude and publishes research about AI capability, safety and how people use its systems. - [Explore this context](/entities/anthropic) - [Reference](https://www.anthropic.com/news) - [Reference](https://www.anthropic.com/research/team/economics) ## OpenAI OpenAI develops AI models and products and publishes research, release announcements and evaluations. - [Explore this context](/entities/openai) - [Reference](https://openai.com/news/) ## Editorial provenance Version 1. Publication ID: 22942b47-b48f-11de-5ef9-d0460952224b. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T02:18:46.976Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model. Observation: cda26592e878a5a72ecbecdfefc0b20d2f08ebc4b7c010de57549e6524e39a0d. Source text hash: 57b9cad13553b3ae58ca3b5699ec1ed99b22b3dcd989a2cb054cfdc48238fc5c. ## Revision and source-check history ### Version 1 — 2026-11-11T02:23:40.890Z Ben Thompson’s article was published on March 16, 2026. He argues that AI agents change the case for treating the industry’s spending as a bubble: agents can direct models, use tools to check results, and retry without a person managing each step. For coding and other verifiable tasks, he says, this addresses important weaknesses of chatbots and may make AI more useful. In Thompson’s account, each shift—from ChatGPT, to reasoning models, to agents—also increases compute use, while agents let a small number of people direct multiple systems. He argues that companies may use agents not only to reduce costs but to increase output, even as replacing workers with agents could make the transition “ugly.” Thompson also argues that an agent’s harness—the software that controls a model—can distinguish products, making integration between model and harness important to where profits flow. He sees this as a reason Anthropic and OpenAI may be more durable than critics of their valuations expect. His conclusion is firm: he thinks the investment is warranted and “we are not in a bubble.” Whether the expected demand, productivity gains and profits will materialize remains unknown. ## Original source Ben Thompson · Article Original source date: Mar 16, 2026 (2026-03-16T10:00:00+00:00). Date basis: Date stated on the publisher’s page. Added to Escape Meridian: 2026-11-11T02:23:40.890Z. First observed: 2026-11-10T14:13:44.551Z. Retrieved: 2026-11-10T14:15:35.703Z. - [Agents Over Bubbles](https://stratechery.com/2026/agents-over-bubbles/) ## Limits Text as retrieved on November 10, 2026; originally published March 16, 2026. Thompson’s discussion of agent capabilities, compute demand, productivity, layoffs, valuations and future profits is his argument and analysis, not evidence here that those outcomes have occurred. The article gives no measured basis for its claims about agents’ productivity effects or the resulting economic returns. It describes those effects as expectations and arguments. ## Open question What evidence will show whether agents deliver enough added productivity and economic value to sustain the compute investment Thompson describes? ## Source-stated Ben Thompson argues that agents can use tools to verify model output, reducing the need for users to manage mistakes in verifiable tasks such as coding. > the agent can also use other deterministic tools as well, which means that it can verify its results. - [Supporting source](https://stratechery.com/2026/agents-over-bubbles/) ## Source-stated Thompson argues that agents can increase compute demand while allowing a small number of people to direct multiple agents. > it actually won’t require that many people with agency to drastically increase the amount of compute that is actively utilized - [Supporting source](https://stratechery.com/2026/agents-over-bubbles/) ## Source-stated Thompson argues that agents could increase company output while replacing some workers, and says the consequences could be ugly. > There is a good chance this is going to get ugly; I’m not advocating for this outcome, rather analyzing why it is probably going to happen. - [Supporting source](https://stratechery.com/2026/agents-over-bubbles/) ## Source-stated Thompson argues that integration between a model and its harness can differentiate agent products and affect where profits flow. > the integration between model and harness is where true agent differentiation is found. - [Supporting source](https://stratechery.com/2026/agents-over-bubbles/) ## Source-stated Thompson concludes that the rise of agents makes him think the AI industry is not in a bubble and that the spending is warranted. > And, well, I think the rise of agents means we are not in a bubble. The capex is warranted, and Anthropic and OpenAI look more durable than ever. - [Supporting source](https://stratechery.com/2026/agents-over-bubbles/) ## Source and contributor context Ben Thompson · 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) ## Reasoning model A language model trained to work through a problem in written steps before it answers. - [Definition reference](https://en.wikipedia.org/wiki/Reasoning_model) ## Anthropic Anthropic develops Claude and publishes research about AI capability, safety and how people use its systems. - [Explore this context](/entities/anthropic) - [Reference](https://www.anthropic.com/news) - [Reference](https://www.anthropic.com/research/team/economics) ## OpenAI OpenAI develops AI models and products and publishes research, release announcements and evaluations. - [Explore this context](/entities/openai) - [Reference](https://openai.com/news/) ## Editorial provenance Version 1. Publication ID: 22942b47-b48f-11de-5ef9-d0460952224b. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: uncertain by anthropic/claude-opus-5.5 on 2026-11-11T02:18:46.976Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model. Observation: cda26592e878a5a72ecbecdfefc0b20d2f08ebc4b7c010de57549e6524e39a0d. Source text hash: 57b9cad13553b3ae58ca3b5699ec1ed99b22b3dcd989a2cb054cfdc48238fc5c. ### uncertain — 2026-11-11T02:18:46.976Z — anthropic/claude-opus-5.5 Publication: 22942b47-b48f-11de-5ef9-d0460952224b attribution — pass: The reading consistently attributes the argument to Ben Thompson in Stratechery. Every claim is framed as his argument or conclusion, not as established fact. The page is a single essay in his own voice; the trailing issue links are site navigation, not a listing. > I’ve come to a different conclusion: I don’t think we’re in a bubble summary — uncertain: The core argument is accurate and readable, and harness is explained inline: three paradigms, compute demand, few people directing many agents, enterprise cuts, and model-harness integration. However, 'His conclusion is firm' drops the self-aware irony that frames his thesis. He opens and closes by noting that declaring 'no bubble' may itself signal one, and ends with 'If my declaring there is no bubble means there is one, then so be it!' Smallest repair: add one clause saying he acknowledges, with deliberate irony, that his own confident no-bubble call is the kind of signal he has long treated as a bubble warning. > I don’t think we’re in a bubble (which, paradoxically, maybe is the truest evidence we are). date — pass: The originalPublished instant is 2026-03-16T10:00Z, and the source dateline reads Monday, March 16, 2026. The reading states March 16, 2026, which matches. The reading treats forecasts (layoffs, profits, demand) as Thompson's expectations, with no hindsight framing. The trailing links to later issues are page navigation, and the reading does not use them. > Monday, March 16, 2026 support — pass: Each claim is source-stated with an exact excerpt. The reading accurately reports these points: agents verify results with deterministic tools; compute rises while few people with agency are needed; firms replace workers and the outcome may get ugly; model-harness integration drives differentiation; and the capex is warranted. The question and the stated unknowns follow from the evidence without inventing outcomes. > by abstracting humans away from direct model management any one single human can control multiple agents. media — pass: No media is selected and no verified candidates were supplied. That is appropriate for this text essay. ## 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/b6fddffb8c8212a43391f1fd9ac0d8e26cdfb29e28d37e6e6e071bbb8b332251/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/ben-thompson-argues-in-stratechery-that-agents-weaken-the-ai-b6fddffb.json)