EXPERIMENT · V0

Try Marcus

Tell Marcus about a problem your team is working through. Explore how companies handle similar situations using research published on HTDI, with links to the relevant primary sources. Marcus is experimental; answers may be wrong, so check the sources. It is open to every HTDI account—no extra permission request needed. If you are not signed in, sign in and return to start the conversation.

How can you try Marcus?

Use the button below. If you are not signed in, sign in to your HTDI account first; the Marcus conversation will then open automatically. No access request is needed.
Type what you are curious about in the chat and send it, or choose one of the suggested questions. For example: “How do companies organize remote work?”
Read Marcus’s answer and open the HTDI pages and primary-source links it provides to explore the examples. Ask a follow-up question in the chat to narrow the topic.

Why this experiment?

HTDI is a very content-dense site. When someone brings us a problem, we don’t say “go read this handbook”; we first understand their situation, then show and compare how companies solved the same problem. What we want to learn: can we offer this kind of consultation in a chat, while staying faithful to the site’s own research?

How does it work?

01
It listens to your problem and, when needed, asks one or two short questions such as team size or way of working.
02
It searches the HTDI library: only company practices, profiles and research notes published on the site.
03
It gives concrete examples from two or three companies, compares them and says which one looks closer to your situation.
04
Below each answer it lists the pages it looked at, and it links primary sources in the text.

What doesn’t it do?

It tries not to talk about companies from general internet knowledge outside the library.
It does not give legal, tax, medical or personal-crisis advice.
It does not speak for companies; it relays research we compiled from documents they published.

Where we may get it wrong

Marcus runs on an AI model (Anthropic’s Claude) and follows strict rules, but it can still make mistakes. The weak spots we know about:

01It can misstate a source
It may summarize a practice incompletely, mix up two companies’ practices, or add a detail the source does not contain.
02It can miss things
Search is still keyword-based. It may miss a topic the library covers, or surface an area that is only loosely related.
03It can misread your situation
The “this one fits you better” judgment is an opinion, not a prescription. It may not know an important part of your context.
04Sources can go stale
Companies change their policies. We rescan our research regularly, but a source may change before we update it.
05It can be inconsistent and slow
It may answer the same question differently at different times. An answer can take 20–30 seconds, and under load it may not answer at all.
06It can link the wrong page
Only site links that come from our research are clickable; still, check that a link matches the claim it sits next to.

Your conversations

Your messages are sent to Anthropic’s API so an answer can be generated.
We do not store whole conversations in our own database; a conversation stays in your browser tab only while the tab is open.
When you copy an answer or rate it as helpful or not helpful, we save that answer, your question, the earlier messages of the conversation and any comment you add, linked to your account. Our team reads them to improve Marcus. If you delete your account, these records are deleted too.
Beyond that, to understand usage we measure events such as opening the panel, sending a message, copying and rating, not message content.
Please do not share personal or confidential details such as employee names or salaries.

Help us shape this experiment

Did Marcus get something wrong, was it useful, did it fall short? Tell us through the button in the corner of the page. This feedback will shape where the experiment goes next.