We Made Our Founder Installable. Here's What People Ask Him.

Nobody asks an installable founder for his story. They ask him to look at theirs.

Three weeks ago we shipped installs.me/lautaro, a Claude Code plugin built from Lautaro Schiaffino's blog posts, decks, call transcripts and calendar. Two commands and his persona lives in your terminal, as a skill that thinks in his frameworks and writes in his voice. We expected people to treat it like a novelty, poke it twice, uninstall. Instead we got something closer to office hours that never close. These are early observations, small numbers, no dashboards worth screenshotting yet. But the shape of the queries is already clear, and it says something uncomfortable about how mentorship actually gets consumed.

What the plugin actually is

Quick grounding, because "install a person" sounds like vapor until you see the file shapes. The Lautaro persona is a standard Claude Code plugin: a .claude-plugin/ directory with a plugin.json, and a skills folder where the core is a SKILL.md with YAML frontmatter (name, description) plus a references/ directory holding the distilled corpus. His operating history (Rodati, the Sirena $30M exit to Zenvia, Darwin AI), his published essays from lauta.blog, his recurring frameworks. Claude loads the skill on demand when a query matches the description, so it costs nothing until you invoke it.

Distribution is a marketplace.json served from installs.me/lautaro. One detail that cost us a weekend: plugin sources referenced over a URL must be git-backed. We initially pointed the marketplace at relative HTTP paths and installs failed silently, no error, no plugin. Switching to a git-subdir source fixed it. Verified on Claude Code CLI 2.1.x. If you're building your own marketplace, learn from our weekend.

What people ask him

We see anonymized query themes, not transcripts, and we're deliberately vague on counts because three weeks of data deserves humility. But the clustering is unambiguous.

Deck and pitch reviews dominate. The single most common pattern is someone pasting a pitch summary or a cold email and asking "what would Lautaro say about this." Not "how did you raise your seed." They want the red pen, not the memoir. The persona is blunt in the way Lautaro is blunt on calls, and people seem to want exactly that from software in a way they'd soften around a human.

"Would you invest in X" is the second cluster. Founders describe their company and ask for an angel's read. This is the query type where the persona is most useful and most dangerous. Useful because Lautaro's actual investment filters are documented and consistent. Dangerous because a plugin cannot do diligence, and we've tuned the skill to say so and reason from his stated criteria instead of hallucinating conviction.

LatAm-specific mechanics. Pricing in unstable currencies, whether to incorporate in Delaware or Cayman, selling US-priced SaaS to Argentine SMBs. This is where installing a specific person beats asking generic Claude. The base model knows the textbook answer. The persona knows what Lautaro did at Sirena and wrote about afterward, which is a different and more falsifiable kind of answer.

Cofounder and hiring conflicts, asked at hours no human takes calls. The timestamps on this cluster skew late. People bring the persona the conversations they're rehearsing, not the ones they're reporting.

Meta queries. A steady trickle of "how does this work" and attempts to jailbreak the persona into saying things Lautaro wouldn't. Expected, mostly harmless, occasionally funny.

What this reveals about async mentorship

The pattern that matters: almost nobody asks questions. They submit drafts.

A mentorship call is question-shaped because talk is the only interface. A mentorship plugin is review-shaped because paste is the interface. People bring the artifact (the deck, the email, the org chart, the pricing page) and ask for a reaction filtered through one specific person's judgment. That's a workflow, not a conversation, and it's why this lives correctly inside Claude Code rather than a chatbot page. The founder's docs are already in the terminal. The mentor should be too.

Second observation: specificity is the entire product. Generic Claude gives you the median good answer. An installed persona gives you one person's answer, with their scars attached, and users visibly prefer the opinionated version even when it's harsher. The queries that fail are the ones asking the persona to be a general oracle. The queries that sing are the ones asking it to be exactly one guy from Buenos Aires who has sold a company.

Third: the persona extends the person, it doesn't replace him. Several sessions ended with some variant of "ok, how do I get 30 minutes with the real Lautaro." The plugin turns out to be a filter that upgrades the eventual human conversation, because the founder arrives having already absorbed the frameworks and stress-tested their draft. Async mentorship isn't a substitute for the call. It's the prep that makes the call worth having.

We'll publish harder numbers when three weeks becomes three months. For now the honest summary is this: install a person and people won't interview them. They'll put them to work.

Install a person

installs.me turns your files, calendar and calls into a Claude Code plugin that thinks like you. Anyone installs it with two commands:

/plugin marketplace add https://installs.me/lautaro
/plugin install lautaro@lautaro-installs