Geoff Woo: GPT-6 Will Allow Ambitious Founders to Announce Company Formation Before Knowing What They Sell
Geoff Woo states that founders will announce before defining.
Investor Geoff Woo claims that GPT-6 will enable ambitious founders to announce the formation of their company even before knowing what they will sell. He also manages AntiFund and has long observed teams that stress-test roadmaps using models.
This statement comes right after the release of GPT-6 Astra. OpenAI claims that this model has reached new levels in controlling computers and browsers, writing software, and mathematics, initially opening to the Daybreak enterprise testing group and paid tiers. President Greg Brockman told reporters that looking back, the time when general artificial intelligence was created might fall around this generation.
Woo's own screening criteria have shifted towards models rather than focus groups: he trusts founders who stress-test roadmaps using GPT-6 or similar systems more than those who conduct product interviews with a dozen people. He only asks one question—are you creating something that makes agents more valuable, or something that agents will bypass? Another record is more specific: the roadshow describes it as "AI operating systems for fieldwork," while the actual business is about eliminating the need for employees to drive forty minutes to fill out a form named pump_status_monday.xlsx; the former raises funds at a thirty times revenue valuation, while the latter makes money.
YC concurrently turns models into equity. Sam Altman offers about $2 million in OpenAI credits to approximately 169 companies in the season, in exchange for unlimited SAFE, converting to equity in the next round of priced financing. He calls it tokenmaxxing: how it operates internally and what products it presents externally are all built from credits. The cost of announcing a company has dropped from renting office space and writing business plans to generating a website, roadshow scripts, and registration documents in one go.
This will amplify shell company launches. Presentations, terms, brand names, and job postings can be completed in one night, while the sales targets and gross profit structure remain blank. Investors face a timeline of "first the entity, then the product," with due diligence needing to shift from product-driven backtracking to agent capability-driven backtracking.
In market mechanics, buyers are founders who must occupy the narrative during the window period and follow-on funds, while sellers are token and accelerator slots from model vendors. Funding shifts from seed rounds that validate demand to pre-seed rounds that validate "announcement capability." Beneficiaries are platforms selling computing power and templates; those under pressure are traditional checks that disburse funds based on milestones and require clarity on what is being sold. The event-driven aspect comes from computer control models compressing the time to generate company shells to nearly zero.
Additionally, Woo did not provide the number of shell launches that have occurred, and OpenAI has not disclosed the timeline for GPT-6 for free users.
Source: Public Information
ABAB AI Insight
Woo is not praising speed; he is describing how the dislocation between valuation and substance can now occur before the registration date. Checks like AntiFund look at whether founders will use models to self-verify; most follow-on investors focus on the announcement itself. During the GPT-3 to Copilot era, entrepreneurship was about "first creating the minimum product and then raising funds"; with GPT-6 being able to control browsers and modify code, the minimum product can be a dynamic slide deck. Historical parallels include how in the 2010s, "Uber for X" could raise seed funding with just a one-page PPT, and how SPACs in 2020 allowed unprofitable stories to go public directly—this time, even the PPT might not be written by a human.
The capital pathway has been rewritten by token investments. Altman’s allocation of credits to an entire YC class effectively turns OpenAI into an invisible infrastructure shareholder for all new companies. Money no longer first becomes engineer salaries but instead becomes fuel for generating company shells. The motivation is to lock in distribution: whoever's model can create a company overnight will extract the default operating system tax for the next generation of entrepreneurship. Woo's question about whether agents can bypass is the only filter that can still separate shells from businesses.
The analogy is not to traditional accelerator Demo Days, but to how app stores once allowed developers to list before finding demand. Stripe Atlas has sped up company registration, and this time, even "what the company does" can be deferred. The industry is transitioning from product-market fit to narrative-market fit. Those who survive will be teams that are immediately pressed by models asking "who is the customer, who is the invoice addressed to" after the announcement.
Structural judgments belong to technological substitution. The mechanism is: when the text, code, and presentations for forming a company no longer constitute constraints, scarcity shifts from execution speed to demand authenticity. The financing market will temporarily give a "first announce" premium until a batch of shells collapses in the next round of due diligence. Pricing power shifts from founders who can tell stories to checks that can identify when stories do not yet correspond to products.
ABAB News · Cognitive Laws
Announcing becomes cheaper, while what is sold becomes more expensive.
Models can create companies but cannot create customers.
First the entity, then the product; the next round will settle the accounts.