Cast Diaries

Notes from backstage: the science of talking to synthetic users.

We read the research so you can decide what to trust. Every claim links to its source, including the studies that disagree with us.

A smiling man in a beanie seen from the side, overlaid with shifted, faded copies of himself and outlined by a hand-drawn red line

Featured · FoundersWhy we needed synthetic users and invented FuzzinessAfter years of chasing customers to interview across consulting, founding and investing, I stopped accepting that research has to be slow and expensive. This is the story of Cast, and of the idea at its heart.April 5, 2026 · 9 min read

A smiling woman at a laptop by the window of a calm, private office

Inside CastYour brief stays yours, and every AI user is labeledTwo promises that shape how Cast is built: your ideas are private to your workspace, and synthetic people are never passed off as real ones.October 13, 2026 · 6 min read

People in a bright open office each on their own headset call

Inside CastTwelve interviews at once: why parallel beats a panelA focus group puts eight people in one room and lets the loudest voice win. Cast runs every conversation one to one, at the same time, and hands you findings in about an hour.September 30, 2026 · 6 min read

A skeptical man with arms crossed in a meeting room, colleagues behind him

Inside CastSkeptics by design: why your cast never all agreesLeft alone, AI respondents drift toward polite agreement and a narrow middle. Cast fights both with deliberate skeptics and a dial we call Fuzziness.September 17, 2026 · 6 min read

An editor with glasses marking a printed manuscript with a red pencil

Inside CastThe Critic who reads every lineEvery answer in a Cast study is checked by a separate AI before it counts. Here is what the Critic looks for, why it is a different model, and what it catches.September 4, 2026 · 7 min read

A young journalist taking notes among stacks of newspapers and books in a library

Inside CastGrounded, with sources: why our AI users do their homeworkAn AI user who invents the price of diesel is worse than useless. Here is how retrieval keeps Cast’s people anchored to the real world, and how you can check every fact.August 21, 2026 · 7 min read

A black-and-white 1960s psychology lab, two researchers in suits working at a control panel

The scienceFrom Milgram to the Ultimatum Game: replaying psychology’s greatest hits on a machineResearchers have re-run famous human experiments with AI participants. Most results come back. The ones that don’t are just as instructive.August 8, 2026 · 7 min read

A smiling young scientist holding up a printed chart in a bright laboratory, colleagues behind her

The scienceCan AI predict the result of an experiment before you run it?A Stanford-led team tested GPT-4 against 70 real experiments with 119,330 participants. The answer, published in Nature, is a careful yes.July 26, 2026 · 7 min read

Two women in a cozy living room interview, one talking with her hands while the other takes notes

The scienceTwo hours of conversation, 1,052 people: why interviews beat demographicsStanford researchers built AI agents of real people from interviews. The agents answered like their humans almost as reliably as the humans answered like themselves.July 13, 2026 · 7 min read

A researcher at her desk late at night, frowning at survey printouts under a lamp

TrustWhere synthetic users go wrong, and how we design around itAI respondents can be too agreeable, too average and too sensitive to wording. Pretending otherwise is how you get bad research. Here is our playbook.June 30, 2026 · 7 min read

A founder at night looking skeptically at a glowing laptop screen

FoundersWhy you can’t just ask ChatGPT or Claude to be your customerThe most common question I get about Cast is a fair one: why not just open a chat assistant and ask it to pretend? Here is what a single chat cannot do, and why the process matters more than the prompt.June 21, 2026 · 6 min read

Actors rehearsing around a table on an empty theater stage in front of a red curtain

Inside CastWhy Cast uses four AIs instead of oneA model grading its own homework is a bad idea. Research on debate, critique and LLM judges explains why Cast splits the work between a Casting Agent, a Director, a Cast Member and a Critic.June 17, 2026 · 7 min read

A woman in a supermarket aisle holding a product and weighing the price

Market researchCan an AI tell you what customers will pay? The honest answerPricing is where synthetic research is most tempting and most dangerous. Here is what the marketing science says, including the parts vendors leave out.June 4, 2026 · 7 min read

Two economists laughing in front of a whiteboard covered in notes

The scienceMeet Homo silicus: what happens when economists interview a language modelAn MIT-trained economist rebuilt classic behavioral experiments with AI participants. The results were familiar, and the surprises were useful.May 21, 2026 · 8 min read

A startup team celebrating with raised arms around a laptop, sticky notes on the wall

FoundersThe fastest way to fail is to build firstMost startups don’t die from bad code. They die from building something nobody needed. Here is what the data says, and why speed of learning wins.May 8, 2026 · 7 min read

Friends laughing together at a lively cafe party under string lights

The scienceSmallville: the AI town that threw a Valentine’s Day partyTwenty-five AI characters, one idea seeded in one head, and a party that half the town remembered to attend. Why this playful paper matters for research.April 25, 2026 · 7 min read

A crowd of people of all ages crossing a sunny city street, seen from above

The scienceOut of one, many: the paper that made “silicon samples” respectableIn 2023 a team at Brigham Young asked a language model to vote. What came back changed how social scientists think about AI respondents.April 12, 2026 · 8 min read