20VC with Harry StebbingsCognition vs Factory | Anthropic Under Threat | ElevenLabs Doubles Its Valuation to $22B
20VC with Harry Stebbings
OpenAI Regains Ground 1:30
OpenAI is said to be nearing a $70 billion run rate and a $1.4 trillion valuation, with an IPO pushed to 2027. The discussion says the pace of change has been sharp, because OpenAI moved from looking behind Anthropic earlier in the year to being back in the race with models that are better and cheaper. The enterprise side matters too, since big customers can take time to switch once contracts, training, and workflows are in place.
Developer Loyalty And Costs 4:31
The group says developers switch tools fast and often use several at once, so loyalty is thin. Open weights models face hurdles in enterprise because of procurement, token costs, token budgets, and IP rights around training data and ownership. That makes the choice of model more than a product call; it is also a question of cost and control.
Factory And Cognition Dispute 10:01
The conversation turns to the dispute over a sales leader moving from Factory to Cognition after serving as a board observer and adviser at Factory. The concern is that someone with access to a founder’s plans can damage trust by joining a direct rival. Others say CROs change jobs quickly and may not see the move as a conflict, but the tension is sharper when the person had real visibility into product, financial, and competitive plans.
Employee moves 15:00
The discussion turns to loyalty when people leave for a rival. A full-time employee can know a lot about your plans and still give notice only after taking another offer, so the speaker says this kind of thing will happen. He draws a line between that and a part-time adviser, where the rules and expectations may be less clear and should have been made explicit.
Recruiting and rivalry 18:30
The group says the AI world has changed how people move, with executives and teams shifting more quickly between companies. Even so, the speaker argues that direct competitor moves still sting, especially when someone has recruited others and then leaves. He says the reputational damage can matter as much as the lost knowledge.
Enterprise demand gap 29:30
The speakers say enterprise use still lags far behind model capability. Costs are falling, and that makes teams more willing to deploy AI across more jobs once they gain experience. They see a large market where both frontier labs and open source can win, so they do not treat it as zero sum.
Safety and model choice 30:00
They also say Dreamforce showed huge internal demand for tokens, with companies overwhelmed no matter how they manage usage. That pressure, plus worries about China-based models, could push buyers toward US options. They are more skeptical of published evals now, since clean benchmark scores do not always match real workflow performance or mission-critical use.
Voice and market research 35:00
The discussion then turns to 11Labs and Listen Labs. 11Labs is said to have doubled its valuation to $22 billion, with strong margins, submillisecond voice response, and wide adoption even at high prices. Listen Labs is praised for using LLMs to improve market research by asking better follow-up questions, and the speakers note that Salesforce buying it makes sense as a feature tuck-in, even if it may not move a giant company much.
Selling early or late 44:01
The speakers weigh early exits against holding on for a bigger outcome. They say timing can matter more than people expect, because a company sold in a hot market can look very different from one sold years later. The point is not that you should always sell. It is that you should know whether you are building something generational or taking a strong offer that may never be beaten.
Agents change the market 49:01
The talk then turns to agents choosing vendors on their own. Versel is used as the clearest example, with agents driving half of new business and creating a strong pull toward its platform. That shifts the problem for every company. You now have to be visible to machines, not just to human buyers, and you may need tools that help shape the content agents read, not only measure it.
LLM marketing upside 58:01
The discussion returns to a new market around helping companies show up in large language model results. The point is that the first pain is discovery, but once you solve that, adjacent problems follow. Marketing teams already have a long list of issues around how they are seen in this new world, and there is money to be made by solving them one by one.
Deal structures and lawsuits 59:30
The talk then shifts to the lawsuit over Gro engineers and Nvidia’s license-and-hire deal. The claim is that a company can be hollowed out through a quick license, cash, and stock package that leaves common holders behind. That raises basic questions about equal treatment under Delaware law, and it may force founders and employees to think more carefully about what happens to equity when only the IP and part of the team are bought.
OpenAI’s product split 1:05:31
The last stretch compares OpenAI’s Muse and Dots. Muse is described as a strong launch with real product pull, backed by compute and clear execution. Dots gets a cooler reception because it is harder to grasp as a consumer product, but it may still matter as a persistent coding agent built on Codex. The main worry is that OpenAI is trying to serve developers, enterprises, and consumers at once, and doing more than one of those well is very hard.
Why The Deal Fell Through 1:11:30
They say the deal likely failed because of price. The board may have been told a higher number by bankers, then pulled back when the real offer came in lower. They also note that selling an entire position would have looked bad, and say the company may simply have chosen not to take a weaker bid.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
