Evidence through late 2025 indicates that startups have not, in general, been able to fully and reliably replace entire functional teams (sales, marketing, engineering) with autonomous AI systems, and that such capability remains years away, consistent with Sacks’s April 2023 prediction.

Key points:

1. AI agents still struggle with real jobs. A 2025 benchmark from the Center for AI Safety and Scale AI tested frontier agents on 240 real freelance projects (software, design, etc.). Even the best agent automated only about 1–2.5% of the economic value, with frequent failures on complex, long-running tasks—hardly a drop‑in replacement for full human teams. (techbrew.com)

2. Industry surveys show low adoption of fully autonomous agents. Gartner-related reporting in 2025 finds that only ~15% of IT leaders are even exploring fully autonomous agents; most organizations prefer supervised or “copilot” style AI, and only a small minority expect AI to replace human workers outright in the next 2–4 years. (techradar.com) This indicates that, even 2+ years after April 2023, most firms cannot or will not run whole functions purely with AI.

3. Leaders in AI infrastructure say full automation is very hard. Databricks CEO Ali Ghodsi emphasizes that people underestimate how difficult it is to completely automate tasks; human oversight is still required even where AI agents are used for customer service and HR workflows. (businessinsider.com) OpenAI cofounder Andrej Karpathy likewise argued in 2025 that current agents are not yet effective and may take around a decade to truly work as hoped. (businessinsider.com) These are strong validations of Sacks’s view that team‑level replacement was not just “months” away.

4. Where AI does “replace a team,” it is narrow and recent, not a broad 2023 reality. Blog‑style case studies in mid‑2025 describe a company that “replaced an entire sales team with 12 AI agents,” but this is: (a) a single, highly specific sales function; (b) reported more than two years after Sacks spoke; and (c) still framed around automating repetitive pipeline work rather than running the whole company independently. (agilegrowthlabs.com) This timing aligns with his “years, not months” claim rather than contradicting it.

5. ‘AI‑only startups’ exist, but only as very small, tool‑stack businesses and again appear years later. A 2025 article describes an “AI‑only startup” with no employees where a solo founder uses tools like ChatGPT, AutoGPT, Midjourney, Zapier, etc., for product, marketing, support, and analytics. (medium.com) These are micro‑businesses operated by a founder orchestrating tools—not credible evidence that, by late 2023, typical startups could drop entire sales/marketing/engineering teams and hand them over to fully autonomous AI.

6. Enterprise and analyst consensus still emphasizes hybrid human–AI teams. Multiple 2024–2025 analyses of agentic AI stress that agents will automate parts of jobs and routine decisions, but not remove humans from the loop in the near term; many agentic AI projects are being canceled or overhyped (“agent washing”), underscoring how far reality lags the marketing. (reuters.com)

Taken together, the data show that: (a) within the first “few months” after April 2023, full functional‑team replacement was clearly not achieved; and (b) even by late 2025, such replacement is rare, experimental, and unreliable. That matches Sacks’s assertion that this capability was years away rather than months, so his prediction is best judged as right.