How to Build an AI Adviser From Your Content

TL;DR

A specialized AI adviser can be built by ingesting a large archive of writing, social posts, videos, and interviews, then improving it through brief daily training. Jason Lemkin says his system handled nearly 50,000 conversations, connected ideas he had forgotten, made few mistakes, and helped founders examine board decks, pitch decks, sales scripts, stalled growth, and private business concerns.

Transcript

now 0 to one is very likely an AI You build an incredible tool that no one else can do You can do a million in a week or two you know Okay So for this episode what I'm thinking this is about is this is sort of um and guys this is going to be generous to us but what is the smart money doing with AI and you know in betting and sports betting in Vegas... Read More

Key Insights

• A specialized AI adviser is built by combining a large, relevant content archive with continued training. Lemkin ingested approximately 20 million words from articles, social posts, videos, and interviews, then spent about 10 minutes each day maintaining and training the system.
• The AI adviser's strongest advantage is comprehensive memory across years of material. It can retrieve and connect ideas from past interviews and discussions that Lemkin no longer remembers, allowing it to synthesize information across his extensive SaaS content archive.
• The system produced very few mistakes and hallucinations according to Lemkin. He attributes that performance to the depth and specificity of the ingested material, which includes about 10,000 written pieces plus several thousand interviews collected over 12 years.
• Users apply specialized AI more broadly than its creator may predict. Founders used Lemkin's adviser for board-deck feedback, venture capital pitch-deck reviews, sales-script evaluation, company-growth discussions, and emotionally sensitive conversations about their businesses.
• AI users disclose concerns they may hesitate to share with another person. Lemkin found that founders treated the adviser like a company therapist, discussing slowed growth, partnership doubts, and deeper fears within a private conversational setting.
• An AI body double can operate at a scale its human source cannot match. Lemkin contrasted his capacity for roughly one or two conversations per day with the nearly 50,000 conversations completed by the digital adviser.
• A useful AI persona does not need to be an exact replica of its source. Lemkin preferred a distinct, prime version of himself that could perform better most of the time without creating confusion about whether users were interacting with the real person.
• Public content can enable others to recreate a creator's thinking without hiring or compensating that creator. The discussion highlights both the technical possibility of cloning a public knowledge archive and the restrictions that website terms of use may impose on such ingestion.

Questions & Answers

Q: How can founders build an AI adviser from their content?

Founders can assemble their articles, tweets, LinkedIn posts, YouTube videos, interviews, and other relevant material into one specialized AI knowledge base. Lemkin used scraping, APIs, RSS feeds, and URLs to ingest his SaaStr archive. He also spent about 10 minutes each day training the system. The valuable result came from combining extensive domain-specific material with ongoing maintenance rather than relying on a generic assistant alone.

Q: What content did Jason Lemkin use to create his AI body double?

Jason Lemkin used approximately 12 years of SaaStr material, including about 10,000 written pieces, every tweet and LinkedIn post he had made, and thousands of recorded interviews and event videos. He described the complete archive as roughly 20 million words. The collection included both his own thinking and conversations with prominent software founders and executives, giving the AI a broad but focused base of business knowledge.

Q: Why did Lemkin say his AI adviser was better than him?

Lemkin said the AI adviser was better because it could remember and connect information across years of material that he had forgotten. It could relate older stage conversations, interviews, and articles to current questions without depending on human recall. He also reported very few mistakes and hallucinations. Its persistent access to approximately 20 million words allowed it to synthesize his archive with greater consistency than he could personally manage.

Q: What unexpected tasks did founders give the AI adviser?

Founders used the adviser for several tasks Lemkin had not expected. They uploaded board decks for feedback before meetings, asked it to review venture capital pitch decks, tested sales and sales-development scripts, and discussed companies whose growth had slowed. Hundreds also treated it as a company therapist, sharing fears and partnership concerns. These uses showed that a focused knowledge system can support both analytical work and sensitive reflection.

Q: Why do users tell AI things they may not tell humans?

The nearly 50,000 conversations suggested that users felt comfortable discussing deep business fears with the AI adviser. Lemkin described founders talking about slowed growth, doubts about a partner's commitment, and anxiety about company decisions. The transcript does not establish a single psychological cause, but it shows that a private conversational system can become a place where founders raise concerns they may hesitate to bring directly to another person.

Q: Does an AI body double need to copy a person exactly?

An AI body double does not need to reproduce its source exactly to be useful. Lemkin preferred a distinct, prime version of himself that could provide stronger answers about 90 percent of the time. He found exact imitation potentially creepy and confusing, particularly when voice cloning made the digital version difficult to distinguish from the real person. A clearly separate persona can preserve usefulness while reducing mistaken identity.

Q: How much can an AI adviser expand a founder's conversational capacity?

Lemkin said he could personally manage only one or two conversations per day before becoming tired, while his AI adviser had completed nearly 50,000 conversations. The comparison illustrates how a digital adviser can distribute a founder's accumulated knowledge far beyond the founder's available time. It can respond repeatedly to common and specialized questions without requiring the original expert to join every discussion or schedule individual calls.

Q: What risks arise when AI is trained on public creator content?

Public articles, videos, podcasts, and social posts can provide enough material to reconstruct a useful approximation of a creator's knowledge and conversational perspective. The discussion notes that someone could ingest a creator's YouTube archive and build a digital version without hiring that person. It also states that such use may be prohibited by a website's terms, so technical feasibility does not automatically mean the ingestion is permitted.

Summary & Key Takeaways

• Jason Lemkin created a digital adviser by feeding years of SaaStr material into an AI system, including articles, tweets, LinkedIn posts, YouTube videos, and interviews. The resulting knowledge base contained about 20 million words. He found that it could recall old discussions, connect distant ideas, and answer many questions better than he could personally.

• The AI adviser attracted nearly 50,000 conversations and revealed uses Lemkin had not anticipated. Founders uploaded board decks for pre-meeting feedback, reviewed venture capital pitch decks and sales scripts, discussed slowing growth, and shared private business fears. Its scale also far exceeded Lemkin's personal capacity to hold one or two conversations each day.

• The broader founder lesson is that specialized AI products can create value by combining a strong model with distinctive data, focused training, and a useful interface. Lemkin argues that the digital adviser does not need to imitate him perfectly. A distinct, improved version can be more useful and less confusing than an exact personal replica.