Why Ambitious Startup Ideas Are Actually Easier To Sell
In a Nutshell
PostHog pivoted to building AI agents that perform entire product and support jobs—identifying issues across data, writing fixes, and iterating in recursive loops—rather than just providing analytics. More ambitious visions are easier to sell to top investors, recruit for, and market because they focus on massive upside instead of downside risk. The company maintains this ambition by shipping constantly, swapping roles to sustain founder energy, and treating marketing with the same rigor as product.
These notes were generated by AI and may contain inaccuracies.
Investors who are stronger and the type PostHog wants to work with focus on what happens if everything works out. They understand that 99.99% of a startup's upside has not been realized yet. The speaker notes that it is all upside and you cannot win on downside.
PostHog came out of YC Winter 20. The team pivoted multiple times during the batch and hacked together the first version of PostHog just before demo day. PostHog was initially a popular YC developer tool focused on open-source product analytics. The company is now reshaping itself to build AI-native self-driving software.
The platform shift with AI led PostHog to up the ambition of what they are doing. From first principles, the team decided it was fundamentally better to entirely solve someone's job for them rather than giving them a rough idea of what is important. They concluded it was better to be a company that does things for people rather than an analytics company that provides an academic view of performance.
The goal shifted to improving graphs and making them go up, not just telling users that graphs are going up or down. The team has been working on clustering and modeling AI research across all data types they have access to in order to identify problems in the product, understand customer complaints in support, and analyze logs, errors, and session recordings.
PostHog is shipping pull requests to fix simpler engineering tasks. The company is also working on higher-level capabilities that are further out.
The vision involves self-contained recursive loops: taking input data, feeding it to an AI along with rules and policies, having the AI take action, feeding results back into the world, observing impact, and iterating on the loop.
Basic components of this system are now in production as of last week. Additional data types are being integrated, including support tickets, video clips of users interacting with the product, and Slack conversations. A desktop product was recently shipped to capture intent behind product usage.
The team is building a harness that captures all intent data from engineers. The system iterates quickly with engineers to understand why a product exists, rather than starting only with technical requirements. This gives the AI something concrete to optimize toward.
In the early days, the co-founders were actively bad at product. During the pivoting process they used to build products that had never existed before. They later realized they were better at out-competing, building product growth, and executing on marketing. With the arrival of AI, they shifted from replacing bits of existing software to identifying jobs already happening inside companies.
The team feels confident they will achieve product-market fit by accelerating jobs that are already occurring inside companies and performing those jobs to a higher standard. A product manager usually decides what to build. PostHog believes a computer can do a superhuman job of understanding everything happening in the customer base across sales calls, internal meetings, customer emails, and user behavior.
There is no reason a system cannot be built that is far superior to a human product manager on the specific part of the product lifecycle that involves synthesizing all customer and behavioral signals.
The nature of the product manager role changes to configuring the software, setting rules and policies, and defining how the system should work. A support product was recently shipped and has about 1,100 companies using it. With access to all the data, the system can solve many support tickets better because it understands the user in more depth than anyone else can.
The support team is not being replaced. Their role is becoming one of supporting the agents: writing the manual and rule book, iterating and improving the system, and essentially becoming philosophers for support.
In Slack, people frequently note problems that already have a pull request ready to merge. The system repeatedly identifies problems and entirely fixes them. A chunk of PostHog's own pull requests are now generated this way, freeing developers to spend more time on big features, new services, and new products.
Trace data for AI-first products allows faster iteration because changes can be backward-tested without waiting for the next human to appear. This capability is already showing promise.
A concern is that the AI could build a thousand tiny features that do not fit together into a holistic product vision. The intent side and the quality of the harness are critical to addressing this.
Current harnesses are not good. When asked to build a to-do list app, Claude will ask technical questions about backend or database but will not ask deeper product questions. In contrast, human product engineers at PostHog will ask why the product should exist, whether it should be simpler or more powerful, who the audience is, and what makes it special.
No one in the industry is currently focused much on grabbing the product side of product engineering. PostHog is working on a broader type of harness. Different modes will likely exist depending on the type of task.
The engineer and CTO jobs have been automated to a degree, but the CPO or product manager layer is still needed to provide product vision and determine whether work is out of scope or off-track.
The decision to make this change was made about a year and a half ago. The idea did not originate from inside PostHog. While on the longest break the founder had taken at PostHog, he read about why some OpenAI co-founders joined and began thinking about the difference between an LLM and a human brain. He concluded there is still a long way to go and that the main constraint on economic value is gluing the model to the use case rather than raw model capability.
Models such as GLM 5.2 are already adequate for many internal tasks. The challenge is making the model actually do the work internally by giving it tools to send emails, run analytics, instrument products, and collect data.
The company is run by two co-founders who consider themselves co-CEOs. The founder with a mixed background (sales team experience and some engineering) swapped roles with the more technical co-founder. The technical co-founder is stronger at running sales as a system, while the other founder is stronger at pushing product direction. They now each focus only on the work they are best at and most enjoy.
When shifting the company toward AI, the two co-founders decided they should not both attempt the transition simultaneously. One co-founder took responsibility for finance, hiring, and scaling the company (which was more than doubling in size that year), while the other focused exclusively on AI and working directly with engineers.
Swapping roles created renewed energy because each person was essentially starting a new job. They observed that when you are not doing something you often start worrying about it, and swapping restored control and motivation. The advice given is not to be a lemming and follow the expected career path of university degree, banking, VP, retirement. Instead, swap roles and run whichever function is currently the bottleneck to company success.
More ambitious ideas are easier to sell to investors, easier to recruit for, and easier to take to market than narrow SaaS tools. Strong people want to work on inspiring work. Ambitious ideas are more remarkable, which helps with word-of-mouth growth in a saturated market. Fundraising is easier for very large ideas because the strongest investors are more likely to engage, even if weaker investors may not.
Founders must hold both the most ambitious version of an idea in their head while also directing the company toward the next practical step that can be taken today. Shipping becomes more important the more ambitious the idea is, because ambitious projects are more likely to fail without continuous real-world learning.
When ideas become more innovative, the first internal question is when the feature can launch and how it can launch sooner. The company is currently under significant pressure to ship because everything has become more innovative.
European founders often wonder how Americans can credibly claim to be working on insane ideas. After going through YC, the founder realized that many YC companies were earlier-stage and more chaotic than expected. Meeting people running $100B+ valuation companies revealed they are ultimately human beings. YC normalized aiming high and set a much higher bar, making the founder 100 times more confident. This culture of declaring ambitious goals (such as solving cancer) is more normal in the US than in Europe.
In San Francisco, hearing extremely ambitious claims like curing all disease is considered normal and plausible. Americans tend to ask "what if it all goes right?" and focus on what would be possible if everything worked. European founders more often focus on downside risks such as what friends or family might think or early regulatory compliance concerns like GDPR.
At Y Combinator, European founders often undergo a kind of cultural deprogramming to shift toward this optimistic mindset.
Startups are default screwed, meaning there is only upside and you cannot win on downside. The investors Y Combinator wants to work with focus on what happens if everything works out because 99% of a startup's upside has not been realized yet. There is no point worrying about downside or how things go wrong.
There is a limit to how much you can earn in a regular job, but running a business can theoretically generate thousands of times more money. The same principle applies to product decisions: focus on upside rather than downside risk.
As an investor, you can only lose your money. Y Combinator has invested in 7,000 companies and estimates that 90% of total returns come from just five companies. It does not matter if 6,995 companies fail if five companies reach $100 billion valuations.
Investing in a company that reaches a $50 million or $100 million outcome is considered a waste of time for investors. The attractive bet is something with a 1% chance of becoming a $10 billion or $100 billion company.
When fundraising, founders should pitch the vision of what could happen if everything goes right rather than emphasizing saving money and being careful. The model investors seek is the high-upside outcome, even when it comes with a high failure rate.
PostHog developed an intentional and distinctive brand, including hedgehogs on buses and campaigns like "you deserve posthog." The approach was deliberate from the start, particularly when entering busy markets where standing out is the primary marketing challenge.
When launching, every competing company had a blue website, used long words, and had pricing pages without actual pricing. PostHog differentiated by being straightforward, allowing users to click sign up, enter a credit card, and start immediately, often with free usage.
PostHog gained significant early traction through non-product factors even when the product was significantly worse than competitors for a long period. People wanted the company to win on the internet. The team targeted engineers rather than other internal roles that competitors pursued.
The product has since become much stronger, though it still has a long way to go.
The team applies the same rigorous thinking to go-to-market strategy as they do to product development. The website is treated as a product internally, considering use cases, target users, and different visitor profiles.
Once this philosophy was established, the marketing team operates with significant autonomy while maintaining a common approach. Most marketing initiatives are run by the team having fun, which produces good results.
Founders and team members genuinely enjoy working at PostHog. The company interviews for cultural fit and seeks people who will be entertained by the work. Candidates showing up in suits or expressing strong desire to be managers are unlikely to be hired.
This enjoyment helps prevent founder burnout over five, ten, or fifteen years. Patrick from Stripe mentioned wanting to run the company for 40 more years, noting that enjoying the role leads to better output for customers.
PostHog pivoted five times before landing on the current product. The team set weekly learning goals such as whether people value a concept, whether they value a particular product, or whether the team can get the product in front of people, rather than focusing on metrics like $20K MRR early on.
The team iterated rapidly toward these goals. Luck in hitting product-market fit is a function of how many shots are taken and the quality of each shot. There are diminishing returns to shot quality; spending 10 years on the first idea would likely still result in failure.
The team would validate ideas by talking to users all day and meeting people in person, even when it required significant travel. For their first paying customer on the fourth idea, they took two or three trains and a bus to meet them for a $300 per month contract, spending roughly that amount on travel.
This in-person commitment allowed them to learn that they could not raise prices later, indicating the customer did not value the product enough. They committed to each idea and hammered at it until they either could not get people to talk to them, use it in production, or pay for it.
Once they repeatedly hit the same hurdle and either ran out of ideas to solve it or felt it did not feel right, they moved on. They were decisive and probably too quick to abandon ideas early because they were initially weak at product development and needed to build that skill through experience.
Keep Y Combinator in your library
Save the videos and channels worth coming back to, and find them again in one place.
