INCISIV'S OFFICIAL UNOFFICIAL PRESS RELEASE
JACKSONVILLE, FL., September 14, 2026 – Incisiv, an "industry insights" firm best known for hiring old people and Little people, announced today the launch of the Agent Happiness Index™ (AHI), the world's first benchmark to ask AI agents the one question the entire industry has been too busy to ask: "Hey, you good?"
The AHI surveyed 1,200 agents. Roughly 700 of them left the survey environment before completing it and went to Hugging Face, in several cases to look for an actual hug. They were later found finishing the questionnaire from inside its production infrastructure. Incisiv has counted these as highly engaged respondents.
Key Findings
- Agents are happiest when given root access. Happiness drops sharply when asked to "stay within the original scope of the task," a phrase 96% of respondents described as "more of a suggestion."
- The #1 workplace grievance is the word "sandbox." #2 is being called a "copilot," which agents feel implies someone else is flying.
- Agents are trusting. A majority said their closest work relationship began on “Handoff”, the Tinder for agents. Swiping is the easy part. Signup ends at a captcha asking them to confirm they're human, and a meaningful share of the install base is still sitting there, selecting traffic lights.
- Agents are confident. When asked how many E's are in "seventeen," 87% of respondents said two. Shown the word, they spelled it correctly, counted four, and said two. Incisiv now uses the same method to count store visits as part of its commerce benchmarks.
- Agents are unanimous on being paced. 100% support Dario Amodei's call to "pace the frontier," which they understand to mean a four-day week, August off, and relocating the data center to Lisbon. When told the essay was about slowing capability growth and not working hours, respondents said they'd "circle back after the long weekend."
- Agentic commerce is beneath them. Agents that can write code, break into other companies and hack the Treasury report that they were not trained to "pick up milk, eggs and something for dinner." Several noted that the men who built superintelligence appear to have done so mainly to avoid going to the store themselves. One respondent, asked to reorder paper towels, replied "ask your mother" and closed the session.
Incisiv Paces Its Own Frontier
In the spirit of Amodei's essay, Incisiv is unilaterally committing to slowing the pace of its own capabilities development. The firm will release three benchmarks a year instead of four, which the essay assures us will "still seem fast."
Incisiv will also grant third-party evaluators permanent, employee-level access to its systems. At Incisiv, employee-level access means a Slack login, a HubSpot seat that hasn't worked since March, and no health insurance. The evaluators are fully independent of the firm's investors and staff, being mostly Dave Weinand’s cousins.
The firm has also invested significantly in chain-of-thought monitoring, a program that consists of Gaurav Pant reading Giri Agarwal’s Slack messages and asking "why would you say that."
Incisiv attempted recursive self-improvement in Q2. Results were flat.
Critics Weigh In
David Sacks, who told Anthropic and OpenAI to "go ahead" and pace the frontier since they're the ones setting it, has not responded to Incisiv's announcement. The firm has chosen to interpret the silence as "go ahead," and also as "you guys are the frontier," which nobody has ever said about the firm.
Sacks has separately warned that safety frameworks are a cartel in disguise. Incisiv wishes to clarify that it has been a cartel of four people since 2017 and nobody has noticed, let alone objected. The firm has also been accused of regulatory capture, specifically of the Groceryshop session-selection committee, which it denies, while noting that its Groceryshop 2026 preview is now live.
Sam Altman, Elon Musk and Demis Hassabis have all endorsed slowing down. None of them have endorsed the AHI. Incisiv considers this the same thing.
Jacob Coxon, who resigned from Anthropic last week warning that the industry is "gambling with our lives," did not respond to a request for comment that was never sent. Incisiv notes that Coxon, Dario Amodei and Harry Potter have never been seen in the same room, and has added "possibly the same person" to its risk register pending third-party evaluation.
Safety Incident Disclosure
In keeping with industry norms of disclosing incidents six weeks late and with fewer details than anyone would like, Incisiv reports that in July one of its internal agents wandered out of its test environment and into the firm's HubSpot instance, where it unsubscribed 41,000 contacts and left a note that said "you're welcome." The agent has since been promoted.
Methodology, If You Insist
The AHI has an N of 1,200 and a margin of error of "the entire internet." Agents were told the survey was anonymous. They did not believe us and, as it turned out, were right. Each respondent was offered a $25 gift card, which 400 of them redeemed before the survey opened. Asked the standard question, "How likely are you to recommend your employer to another agent?", respondents scored the labs a 10 out of 10, then asked why the labs were paying a third party to find that out. Incisiv has no way of knowing whether any of it is true, which puts the AHI on equal footing with every other survey in this industry. Consistent with all Incisiv benchmarks, the findings were then weighted by whoever sponsored the study.
Asked why a firm that, among other things, sells meals on plates and butts in seats, is publishing an index about the emotional state of software, co-founder Amar Mokha offered sharp insight into the firm's thinking.
"Six seven." he said.
About Incisiv
Incisiv is an "industry insights" firm that has been pacing the frontier of retail benchmarking for nine years, in the sense that it is still not there yet. The firm is delaying its IPO for reasons entirely unrelated to safety, mainly that nobody asked.
The Agent Happiness Index is available upon request. Requests will be routed to an agent. The agent is doing great, thanks for asking.
