Everyone is building the same thing

I think the tweet of the week was “Everyone. is. building. the. same. thing.” by Tom Osman. His tweet was vague but it was seen by nearly half a million people and liked 2.3K times. A few days later, in a conversation between Openrouter co-founder Alex Atallah and Replit CEO Amjad Masad, the same concept returned in more tangible detail. Everyone, Atallah said, is building the same AI-stack as they figure out how to build their AI-native applications. “Everybody's building an agent loop with notifications, third-party connectors, context management, memory, sandboxes, agentic web search, an always-on agent on top…This product is showing up everywhere…but these are just the new table-stakes primitives. "A 2005 version of that tweet would be: 'Oh, everybody's building the same thing. A database, a users table, a sign-in page, a sign-up page, a profile page, a logout page. Everything's the same. There's a lot of differentiation really. There's table-stakes needs for AI just like there's table-stakes needs for the web." Alex is saying there IS differentiation, but it’s not where people are looking. More on that below.

As someone who was around for the earliest days of the internet (as an equity research analyst) and for the heyday of “Web 2.0” as a very junior VC, there are absolutely echoes of those eras. (everyone bought servers, then everyone bought cloud, then everyone used the same APIs, etc). With every company and every engineering team rushing to try all the latest approaches and frameworks and needing to build every layer of a new stack that is not yet stabilized, the level of confusion rivals the level of intensity. Today, tech is still evolving too fast for there to be any canonical AI-native stack. There will be, but it's not here yet. And until it arrives, it’s going to be too easy for us all to confuse table-stakes with differentiation.

Everyone is saying the same thing. Not only is everyone building (a version of) the same thing as they struggle to get their AI-native apps off the ground, everyone is communicating about them in the same way using the same tools. A growing number of VCs are now publicly begging founders to stop using AI to put together their pitch decks and materials. Even Sarah Guo, founder of Conviction Ventures, joined the cause this week. She implored founders on X this week to “please stop sending me your ai-generated decks. they smell like total lack of thought” and she’s 100% right. A disconcerting (and growing) percentage of our dealflow is companies that increasingly look nearly exactly the same because they are building the same thing and talking about it the same way. (I’ve written about it before so I won’t do it here, but I could have easily added a paragraph about how everyone is funding the same thing. That, however, is a slightly different topic.)

Same same but different. In that sea of sameness that seems to be the hallmark of our times, there are still some very interesting and unique companies trying to raise capital. Even more interesting, however, is that idea that even within the same category it is still possible to stand out. What we are observing is that the uniqueness is not in the AI-native stack. It’s not in the harness, the context layers, the memory graph, or the evals. And it’s definitely not in the pitch deck. It’s somewhere else. It’s in the deep domain expertise of the team that led them to a problem. It’s in customer empathy and clarity of value prop. In today’s climate, I find myself doing customer references and team interviews with even more intention. The more we converge on a set of AI-native primitives, the more the idiosyncrasies and passions of the founding team matter. And the more AI has automated the build, the more the human-to-human connection between makers and customers matters. The tool stack will keep changing and will - ultimately - not matter a whit. Asking a founding team about their context memory approach, their token efficiency, or which model they are building on today is as silly as asking a team about their log-in page and user database choices in 2005. What is exciting to me when I look at vertical AI-native companies today is the passion they and their customers have for the product and the earned insights that underpin it. That can’t be vibe-coded and that doesn’t depend on a stack that is yet to be standardized.

Gil Dibner

Could the future of software be fluid
How do we get the best of AI without losing the soul of software?

The future belongs to young missionary teams
Why it makes more sense betting on youth in the current moment

The AI-native enterprise playbook
Ten real-time observations on a rapidly evolving playing field

No more painting by numbers
It’s the end of the “SaaS playbook.

HARD TECH

Closer to the metal - Fei-Fei Li explains in her own words why World Labs is joining AMD for $8.2B, two years after founding. Models that reason about the physical world need silicon built for them: "Without having a focused hardware effort, AI is hobbled in efficiency." Her proof point is Atlas, which predicts the next camera view the way an LLM predicts the next token.

HARD MARKETS

Parameters - Bloomberg reports that Schneider Electric is buying industrial software firm PTC for $22.6B in an all-cash deal, its biggest ever, at a premium of more than 40%. Investors balked, sending Schneider down as much as 10% in Paris as analysts flagged the leverage, though Jefferies wrote that the deal "fills one of the remaining gaps in Schneider's software portfolio." The piece also explains why industrial software has dodged the AI sell-off hitting SAP and Salesforce: it runs on proprietary data and processes with no margin for error.

HOW TO STARTUP

Defensibility in AI Data - Gokul Rajaram warns founders selling data and RL environments to AI labs that hypergrowth doesn't guarantee a durable business, pointing to the early-2000s ad networks that boomed and then lost nearly all the value to the exchanges. Today's vendors risk the same fate: "The supplier gets paid and starts looking for the next project." His three ways out are owning differentiated supply, embedding in the lab's workflow, and building environments that improve with every training cycle.

Study your dumb moves - Slava Akhmechet distills what a year of serious Go taught him about competence: you improve by dwelling on bad moves instead of shrugging them off, because they expose faulty instincts, gaps in understanding or the wrong attitude. He sees founders fall into the same trap, sticking with their ideas and hoping instead of talking to users. Go skill didn't transfer to his day job, but learning how to learn did, along with the players' cure for beginner's embarrassment: "lose your first 50 games quickly."

HOW TO VENTURE

Small seeds, big variance - Jack Altman joins Rory O'Driscoll, Jason Lemkin and Harry Stebbings for a 20VC episode last week. Despite headlines around recent mega seed rounds of $50M, they make the case that $2–3M is still the right size for a first round since AI tooling lets a lean team get much further on it than a decade ago. The discipline matters more as outcomes polarise, with Tyler Cowen's warning invoked: "Variance is gonna go up with AI, and many of you will fail." They also expect more AMD-style acquisitions, with few of the roughly 100 independent research labs surviving on their own.

Reco raises $55M with AT&T backing to secure AI agents.

CruxOCM
Forward Deployed Product Manager (Remote, Calgary)

Reco
Product Manager (Tel Aviv)

Sourcix
Full Stack Developer (Tel Aviv)