AI is quietly stealing your life’s work
with Dr. Jonathan Schafer, Kind
AI is quietly stealing your life’s work
Show Notes
Every document, idea, and decade of hard-won knowledge you have uploaded to ChatGPT, Gemini, or Grok - that is not your data anymore. You paid for the privilege of training someone else's billion-dollar model with your life's work. There is no free lunch. Read the terms of service.
Dr. Jonathan Schafer has been working in AI since the late 1970s - long before it was called AI, long before there was a market for it, and long before anyone had heard of a large language model. He is a 40-year professor, an early machine learning researcher who trained over 75 graduate students, and he has built something he calls Kind: a privacy-first AI that runs on your laptop, answers questions only from your data, and never sends a single byte to the cloud.
This episode is about the moat that most founders have overlooked - the unstructured IP locked inside their own machines - and what it means to build AI that actually serves the individual instead of the corporation.
The Data Problem Nobody Is Talking About
When you upload a document to a cloud AI provider, you are entering a transaction. The AI does something useful for you. In return - buried in the terms of service - you have often granted that provider the right to use your data for training, analysis, or resale. For most users, this is invisible. For founders with 40 years of proprietary research, unpublished frameworks, client work, and voice memos, it is a material IP risk.
Jonathan's answer is not to avoid AI. It is to use AI differently. Kind runs entirely on your local machine. Your laptop's idle compute - which is most of the time - handles the analysis. Your data never leaves. There is no account with an LLM provider. There are no hallucinations because the model is not drawing from the broader internet - it is drawing only from what you put in.
The pitch is elegant: you curate the data. Kind curates the AI. A librarian who knows everything in your library, and only your library.
The Unstructured IP Opportunity
Jonathan and Ryan landed on something that deserves its own conversation: the value of unstructured IP that currently lives on laptops, in filing cabinets, and across drives that nobody has opened in years. For a 40-year professor, that means courses, research papers, conference transcripts, student assessments, and voice memos accumulated over a career. For a founder, it might mean deal memos, customer research, internal frameworks, and institutional knowledge that only exists inside one person's head and hard drive.
Until now, that data was inert. It could not be interrogated. It could not answer questions. Kind changes that: drag your files into a collection, let the system do a one-time AI analysis, and suddenly you can ask your data what is there, what has changed, what you might have missed.
Jonathan discovered, using his own product, that he had been underteaching security in his operating systems course - a gap he would never have found by reading the slides himself. The AI found it dispassionately, cited the specific files, and gave him something he could act on. That is the use case: not a generic assistant, but a tool that knows exactly what you know and helps you know it better.
AI for the Individual, Not the Corporation
Jonathan has a clear thesis: all the energy in the AI industry right now is flowing toward corporate efficiency - reduce headcount, compress workflows, improve margins. That is real value, but it is not where AI was supposed to go. AI was supposed to benefit people. The billion individuals who are not enterprise software buyers. The retiree with a lifetime of photographs. The author whose books are being quoted without attribution by models trained on them without consent.
Kind is a bet that there is a market for AI that is genuinely private, genuinely personal, and genuinely useful to people who are not building enterprise pipelines. The hallucination problem goes away when the model has no internet to hallucinate from. The privacy problem goes away when the data never leaves the device. The complexity problem goes away when the user does not need to know what Claude or Gemini or GPT is.
Whether that market is big enough to build a company on is the open question. But the insight - that individual IP is valuable, under-leveraged, and in need of protection - is correct regardless of the answer.
What Roald Amundsen Teaches Founders About Preparation
Jonathan's book recommendation is The South Pole by Roald Amundsen - the Norwegian explorer who beat Robert Falcon Scott to the South Pole in 1911 by about a month. Scott's team froze to death on the way back. Amundsen's team returned without a single casualty. The difference was not luck or conditions. It was preparation. Amundsen had planned for every scenario. Scott had not.
The parallel to company-building is direct: the founders who survive the brutal stretches are not necessarily the ones with the best ideas. They are the ones who thought through the failure modes in advance, built in the contingencies, and moved deliberately through risk rather than hoping for favorable conditions.
Frameworks from This Episode
These frameworks have been added to the AI for Founders Frameworks Library. Filter by Privacy or Jonathan Schafer to find them.
- The Dormant IP Audit - Treat the unstructured data on your hard drive as a product waiting to be interrogated. Decades of research, client work, and institutional knowledge become queryable assets the moment they are fed into a local AI system.
- The No Free Lunch Rule - When an AI platform offers free or low-cost access to powerful models, the implicit transaction is your data. Read the terms. If privacy matters to the use case, the cost of cloud convenience is potentially your IP.
- The Amundsen Preparation Standard - Before undertaking a high-risk venture, systematically think through every failure mode. The explorer who returns safely is rarely the boldest - they are the most prepared.
Tools Mentioned
These tools have been added to the AI for Founders Tools Directory.
- Kind App - synsira.com - Privacy-first AI that runs locally on your machine, answering questions only from your curated data collections. No cloud, no hallucinations, no training on your data.
Glossary
Terms from this episode have been added to the AI for Founders Glossary. Filter by Jonathan Schafer to see them all.
- Local AI - An AI model or system that runs entirely on a user's own hardware, with no data sent to external servers. Eliminates privacy risk and enables hallucination-free responses when the model is constrained to a specific local data set.
- Unstructured IP - Proprietary knowledge, research, client work, and institutional memory that exists in disorganized files - PDFs, voice memos, images, documents - and has not been indexed, analyzed, or made queryable. Represents significant latent value for founders and domain experts.
- Data Collection (Kind) - A curated subset of files on a user's machine, analyzed once by Kind's local AI, that becomes queryable without any cloud access. The unit of organization for privacy-first AI interaction with personal data.
- Hallucination-Free AI - An AI system constrained to answer only from a specified data set, with no access to the broader internet or general training data. Eliminates invented or inaccurate responses by removing the source of uncertainty.
- No Free Lunch (AI Privacy) - The principle that cloud AI platforms offering free or subsidized access to powerful models typically extract value through terms-of-service rights to user data - for training, analysis, or resale. Every convenient upload is an implicit transaction.
- Founder-Led GTM - A go-to-market strategy in which the founder is the primary public voice - doing podcasts, content, and community building rather than delegating outbound to a sales team. Particularly effective in 2025–2026 as authenticity and domain credibility drive customer trust.
Q&A: What Founders Ask After This Episode
Is my data actually being used for training when I upload to ChatGPT or similar tools?
It depends on the platform and the tier. OpenAI's free and some paid tiers have historically used conversation data for training, though enterprise tiers typically opt out. Gemini, Claude, and others have similar tiered policies. The honest answer is: read the current terms for your specific plan. Jonathan's point is that most people do not, and the default assumption that the interaction is private is often wrong.
What kind of data works best in Kind?
Any unstructured files you have accumulated over time: PDFs, Word documents, presentation decks, voice memos, images with metadata, research papers, course materials, client notes. The value is proportional to the depth and specificity of your archive - the more you have built up in a domain, the more useful the interrogation becomes. Jonathan found insights in his own teaching materials he had never noticed after 40 years.
How is Kind different from a local LLM like Ollama or LM Studio?
Local LLMs require technical setup, model selection, and prompt engineering. Kind abstracts all of that away. You do not need to know what a language model is, choose model sizes, or manage infrastructure. You drag your files in, and Kind handles the AI layer. The trade-off is less flexibility for advanced users in exchange for zero technical barrier for everyone else.
Is the unstructured IP on my hard drive actually worth anything?
Jonathan's answer: for domain experts who have spent years or decades accumulating knowledge, the answer is yes - potentially significantly. The data is unique to you. It cannot be replicated. It represents the actual reasoning, judgment calls, and institutional memory behind your expertise. The gap is that it has been inert and un-interrogatable. Tools like Kind convert that latent value into something you can query, summarize, and build products from.
What does the Amundsen principle mean for founding a company?
Amundsen beat Scott to the South Pole by a month and returned alive. Scott's team froze to death. The difference was not talent or ambition - it was that Amundsen had mapped every possible failure scenario and built contingencies for each. Applied to startups: before committing to a market, a product, or a spend decision, systematically think through the failure modes. What breaks if this goes wrong? What is the recovery path? Preparation is not the opposite of speed. It is what makes speed survivable.