Product won't end at chat

Consumer AI, trial by combat, and the vSaaS cycle

AI capabilities have run so far ahead of our ability to use them well, design for/around them, incorporate them into life and business. All the foundation models could stop today and there would still be YEARS of work to do before we hit the limits of a new biz models and UX.

We’re all still stuck trying to reason with and convince our computers to do what we want. This is not where product ends and the limitations on usefulness or not reasoning, memory, etc.

I heard one really cool idea this week: an LLM spinning up new UIs and mini apps within chat based on need. OpenAI is sort of already doing this with maps but this was a much more expansive play.

Chat is awesome for super horizontal use cases but it sucks for depth, workflow, and longitudinal data. There’s a reason why the AI code generators are either set up as IDEs or agents rather than mere text boxes - that is just not useful for more complex tasks, even within the domains that LLMs are best.

In the early internet era, lots of different advertising, media, and retail businesses got bundled together into eBay, Craigslist, etc. Then over the next decades they got unbundled into more focused vertical businesses. Arguably the next platform shift saw operating systems unbundled into apps. Today, lots of different tools are getting bundled back together into horizontal/general purpose chat apps.

It’s happening super quickly but it’s the same cycle. One horizontal platform will not be the steady state or best outcome. The same thing will happen and the AI apps will disaggregate back into deeper apps/tools/businesses that can go beyond chat because they are built to purpose, at least within some categories and behaviors.

There is a huge amount of whitespace to rebuild consumer apps and services with AI at the core, but not as pure chat apps. Even if natural language is a great input, you need pre-made images/videos, time series data and tracking, deterministic workflows, external connectors, etc. to get the most value back; chat alone can’t furnish that.

Anything that has been a guidebook or professional guide (travel, game walkthroughs, fitness, cookbooks, college prep, self help) is a candidate for a standalone consumer AI app.

People pay for this already and horizontal chat apps won't be the right UX. So you need to think beyond chat as the sole interaction. There are incredible new products to build for those with the vision to see them. The fun stuff in consumer is just starting/hasn’t really begun yet and I’m excited to see what people do next.

Here’s one such example:

AI homebuying is an obviously good consumer app

Historically, you needed an agent to access housing inventory, assess a home, negotiate the deal, and close the purchase. This was an important, high touch service and buyers paid dearly for it: ≈3-5% of the purchase price. Obviously every dollar in agent fees is $5 less house you can afford.

That fat upfront fee is now effectively optional after the NAR settled its antitrust case. Buyers and sellers can now negotiate fees with their agents independently or represent themselves.

But even if it's theoretically possible, most people still need/want help: there's been no meaningful increase in self representation and the fees basically haven't moved. This is not surprising. Buying a home is most likely the single biggest financial decision you'll ever make and it's largely a one way door: you're locked in for years with all your new worth tied up in a single, highly levered asset.

We should start seeing AI real estate agents as a lower cost option soon. Search and discovery is already self-service through Zillow and Redfin but now AI can run the homebuying process and save consumers a ton of money/make down payments much more affordable.

The most important parts of self-service AI homebuying experience will be 1) Project management (coordinating vendors/inspection, title, scheduling showings) and 2) Advising on the transaction (generating and negotiating an offer, reading contracts, applying for a mortgage, etc.). This is already well within the bounds of AI capabilities but it needs the right product. Chat won’t cut it.

This is a much better customer acquisition pathway than lots of other home services companies: it's specific moment in time, obvious intent/CTA/value prop, directly attributable savings, and short time to value. If you do it well, you still retain people into all the various home management stuff people want to do with AI (which don’t seem to be working yet)

The optimal pricing model and product for this are both TBD. Figuring that out will be the fun part!

More from me

• The actual vertical AI dynamics aren’t quite the same as last gen vSaaS but the frenzy is. There are some reasons to think this might be meaningfully different, namely: shorter time to value, bigger deal sizes, clearer value props. Small markets become bigger on that basis. But ultimately, the winners will wind up in some kind of multiplayer mode (collaboration or nfx to create retention and margin).

• We can/should/might/will recreate trial by combat with AI to settle more disputes faster. You don’t need to get things “right;” you need fast, definitive answers bound by reasonableness, especially in the low level contractual disputes that gum up the works of commerce and daily life. That is, you need socially legible venues and rituals to confer legitimacy. Asking the LLM = grabbing a hot iron = shaking a cup of animal bones.

I Read

• Meet the 24-year-old Ottawa software engineer who runs a MAGA bot. I thought Elon was getting rid of the bots?? In all seriousness, this is where our politics and online spaces are heading. Truly poisonous stuff.

• The U.S. Marches Toward State Capitalism With American Characteristics. MAGA Maoism: now in the pages of the WSJ. To be clear, eroding the rule of law, undermining markets, and picking winners does not end well. Maybe we need to call it US Peronism?

• Why most VC firms hire associates wrong (and what we do instead) | Amplify Partners. Brutal and accurate assessment of VC associate programs/roles. I don’t think it’s a simple binary between 2 year grunt leverage and long term apprenticeship. We have tried to base our associate model on USV’s where we teach for a couple years and get leverage in exchange. But we don’t retain people past 2-3 years because we're too small and can’t absorb 2-4 new partners every couple years.

• Perplexity / Chrome … and the future of product lead SPACs?… or AOL / Time Warner Calling? This is a great bit/take. I don't think it'll happen but it's an awesome hype generator for Perplexity.

• The Woodworking Creator Who Raised $2 Million This is the first deal from our new $63M creator fund. We backed Jonathan Katz-Moses, a woodworking creator and the founder of KM Tools as his first outside investors. This is exactly the type of founder and story we want for the creator fund: entrepreneurial/hungry and with an audience around a valuable niche.

• Investing for exposure. “It’s not just public markets. LPs & GPs run exposure math, too. An endowment recently told me about how they’re modelling look‑through exposure to foundation models. i.e how many dollars, through layers of VC funds and SPVs, actually touch xAI, Anthropic, OpenAI, SSI, Thinking Machines etc... Company execution? Important, but not the exercise. First: “Do we have enough dollars in the ground?” This is a very Dempsey take and I think it’s right.