GPT-5, the AI bubble, and the Windsurf employee payout scandal

Executive overview

AI model improvements are slowing, not accelerating — each new release is a smaller step than the last. Meanwhile, AI compute is being sold far below cost, creating fragile businesses built on subsidised pricing. Separately, the Windsurf acquisition exposed how founders can capture nearly all value from a deal while employees get nothing.

The social contract of Silicon Valley — where early employees share in an exit — is being systematically dismantled.

Tiny Seed Fund 1 returns capital to investors

• Fund 1 has returned more than 1x DPI (distributions to paid-in capital) to LPs.
• Only ~50% of venture funds return 1x DPI even after 12 years; Tiny Seed did it in six.
• Most fund assets are still growing — DPI is on top of remaining upside.
• Returns are power-law driven: one or two exits or partial exits are responsible for most performance.
• Fund 3 is currently open to accredited investors at tinyseed.com/invest.

GPT-5 and the AI capability asymptote

• Each new model release delivers a smaller capability gain than the previous one — an asymptote, not an exponential takeoff.
• Scaling laws mean each capability step requires exponentially more compute; the "FOOM" scenario is not supported by the data.
• GPT-5 is capable but not superhuman; it does not make existing SaaS obsolete.
• Model specialisation is the real trend: different models optimise for different tasks (conversation, coding, reasoning).
• GPT-4.5 was preferred by some users for conversational quality; its discontinuation frustrated power users.
• Founders should stay current with AI but do not need to abandon their businesses.

The AI bubble and subsidised compute

• AI providers are selling compute at a fraction of its true cost; the gap is funded by venture capital.
• SaaS products built entirely on cheap API access may be unviable once pricing normalises.
• Mitigation options: fine-tune open-source models for your vertical, reduce dependency on frontier API providers.
• Founders building on AI today need to actively plan for the day pricing reflects real costs.
• Smaller, cheaper models (e.g. GPT-mini equivalents) may be "good enough" for most use cases.

The Windsurf debacle: employees shut out of a $2.4 billion deal

• OpenAI was set to acquire Windsurf for $3 billion; the deal collapsed.
• Google instead paid $2.4 billion for a non-exclusive technology licence and hired the CEO, co-founders, and key R&D staff.
• No equity or company control changed hands, so common shareholders and early employees received minimal or no payout.
• A second acquirer (Cognition) picked up what remained of the company.
• The founders and investors captured roughly half of the $2.4 billion; employees did not.
• This is distinct from a soft landing: the deal size was close to the original $3 billion acquisition price.

What early employees can actually do

• Equity outcomes increasingly depend on what founders choose to fight for in deal negotiations.
• Investors and acquirers have accumulated contractual tools (liquidation preferences, retention carve-outs, etc.) that reduce common shareholder payouts.
• Alternatives for employees: join a FAANG-tier company for guaranteed high compensation, or job-hop to increase salary rather than betting on equity.
• The Windsurf case is not a soft-landing story; it is a structurally bad outcome at a large deal size.

Selling when your user is not the buyer

• Most B2B SaaS eventually requires selling to someone other than the end user (e.g. developer uses product, CTO approves budget).
• Build tools and resources that help your user make the case to their decision-maker.
• Sales is understanding what the other person needs and matching it — not a separate skill from product thinking.
• Self-serve models can work to $1–3 million ARR; unlocking enterprise sales is typically required to reach $5–20 million.
• Jason Cohen's "max MRR" concept: given current churn and new MRR bookings, you can calculate where your growth asymptotes — and act before you hit it.
• Founders who avoid enterprise sales often stall, then sell to private equity at a discount, leaving most of the value creation to the acquirer.