Thoughts I've been exploring recently

18 Aug 2026 · Voice AI

A few thoughts on voice AI

Voice AI is getting really good. Actually, it's kind of awesome already. The models are there, but their penetration still seems quite low.

I've tried a few voice AI bots focused on long-horizon interviews — 20 minutes or more — and it is a very distinct experience from answering the same questions in text. It feels like something live. It has more energy and kind of deserves your full attention. In a chat, you can close the window, go do something else and get distracted. A live conversation, even with a chatbot, feels focused.

Because of that, voice AI interviews for research and audits seem very interesting. One of the main parts of any consulting project — and an AI deployment project specifically right now — is auditing how processes actually work: what people dislike, where they see optimization potential, what happened before and which mistakes were made.

Of course, one of the main sources is to deploy agents to analyse all internal knowledge. But internal knowledge is frequently messy and doesn't contain the opinions you need. It may be much faster, more convenient and higher quality to deploy dozens of agents to run live voice interviews with employees and team members.

The same can be done for regular HR research and, of course, customer research. There are companies growing quickly around this, like Listen Labs for customer research and the newly created LATO, founded by the brother of an ElevenLabs co-founder, for commercial due diligence. Customer research looks quite competitive already, but LATO seems uniquely positioned. Curious how it performs.

I believe focusing on a particular professional profile is the way. Plenty of “wrapper” companies that VCs claimed had no moat are thriving, so generalist agents can work. Still, in research, specialization feels key: it lets you automate the full service and build end-to-end software around it.

18 Aug 2026 · Ambient AI

A voice companion for walks

Voice AI is cool. Latency is low, quality is high, and it is actually kind of natural — even sometimes enjoyable — to speak with right now.

There are companies that provide real-world data for AI, and this opens new use cases. For example, a voice companion that activates by itself during your walks. When you are close to something cool, it tells you the history of a building or a story about the street.

Usually this stuff is pre-recorded by museums and brands, which is one way to do it. But AI is good enough now that, with a good knowledge layer and a pipeline that gives you only good facts — in the tone you like and about the topics you like — it could be completely automated and lively.

The devil here is in these small UX details. I like hearing stories about buildings and architecture sometimes, but I really hate taking out my phone, opening it, taking a photo and asking ChatGPT what it is. It would be cool to just have a companion that is always on.

Ideally, it should work even if you're listening to music on your walk. Either you allow it to interrupt the music — with a setting to turn that off — or it listens for your voice and you just say, “Hey, what's this building?” and it answers you. Meta's glasses and similar devices are also a good way to integrate an application like this.

17 Aug 2026 · Consumer AI

What will unlock consumer AI?

While both OpenAI and Gemini reached 1bn MAU, the conversion to paid is around 1.6%. Why is that?

A few hypotheses.

Most consumers don't have problems where AI is a drastically better solution they want to pay for

Would be good to map all B2C spend on existing software and services — AI can automate human-provided services — and see the market segments and what's already there. Then map AI penetration by use case, and the mode of AI: existing app, chat, or new vertical product.

To do — market map

The chat is a bad universal UI — something else will unlock the usage

Chat is bad for discovery: people don't always know what's possible. Chat is a boring UX: people want engaging and case-specific digital environments. Major labs tried app integrations and MCP apps to no success — why? The Wabi founders think apps should become on-demand, a curious idea too, but the greatest apps require mastery and skill.

Consumers do use consumer AI — it just got embedded in the apps they already had

People continue to use the same apps they actually use: calorie trackers, fitness trackers, whatever trackers. AI becomes an embedded, native part of these products — either directly seen (a Copilot, a chatbot) or just behind the scenes, providing new benefits and features.

There are new vertical consumer AI products — just niche?

There are AI friends — Character.AI and the like — that look like real consumer scale. AI for photos and videos seems to be growing big. And there are super popular, high-penetration cases in B2C edtech: AI tutors for English, maths and the rest, embedded in Duolingo, Preply and others.

17 Aug 2026 · What got cheap

What did AI make so cheap that something new becomes possible?

Obviously AI is the biggest inflection point of our generation, but I'm not curious about the general point of view. It's clear how it lets us do things at scale, cheaper. I'm curious about the particular areas where the inflection creates new possibilities and behaviours — something completely new. What got so cheap or so fast that it creates new user experiences?

Games get cheap enough to be used as media

Subscribe to the a.jsx account on Twitter and you'll see that agents are already outstanding at creating procedural 3D worlds and games. Literally, small scenes and small games can be one-shot, which means the production of small, on-demand, personalized worlds and games is close. It's already there with some quality. Top-notch games are works of art, but games of some quality will be easy to produce.

I'm curious about whether this does not just increase the amount of games in the world, but allows us to use games as media in places where they have not been used before.

Each newsletter is a new game.
Gamified on-demand education everywhere, for kids.
Websites.
Any other place where a cheap game is better than text or video.

We don't know. Just curious to think about it.

Human services get automated — the obvious one

Everyone is talking about it. The most direct thing: some part of human-provided services can be automated. In some cases fully, like customer support in high-frequency cases. In some places by augmenting the service provider, as in legal. Is this just the first, most easily accessible idea? Are there any service providers left that are not yet tackled by dozens of competitors?

Which consumer services is nobody serving?

Most AI deployment is focused on B2B services — predictably, because there is more money there. What B2C services are not tackled yet? Why is there no high-quality travel agent, shopping agent, and so on?

Tacit knowledge finally becomes writable

Cheap intelligence allows the formulation of distributed tacit knowledge. It means decision tracking, knowledge, and all the complex, interconnected processes in a company's knowledge graph can be formalized, kept track of, and potentially improved at a scale that was not possible.

Research fleets instead of ten blue links

I personally really like how search, relevance review and quality review can be done with agents at unprecedented scale — instead of us just looking through the first dozen results in Google, or even just believing one agent that does a few dozen searches. You can create research fleets that know precisely the job to be done of your search, and your personal preferences. They can even be programmed not to insulate you in your bias, but to challenge it and always show adversarial counterpoints.

A fleet can search at unprecedented scale, analyzing not dozens or hundreds but thousands and thousands of internal or external knowledge pieces. Technically this allows you to find what you need better and more frequently — and to create personalized newsrooms and media.