Available evidence by roughly one year after the podcast (through late 2025) shows that AI customer service software did not become a fully commoditized, low‑value startup category, even though basic AI features became common.

1. Major AI customer‑support startups are still capturing large, differentiated value. Sierra, an AI customer support startup, reported its ARR growing from about $20M to $100M in a year and was valued around $10B, after raising more than $600M. This is specifically an AI customer‑support company, not a foundation‑model provider, and it is expanding headcount and office footprint rather than being squeezed out by commodity competition. (theverge.com)
2. Incumbent application‑layer vendors doubled down on AI and maintained strong valuations. Intercom, whose core product is an AI‑driven customer service and messaging platform, rolled out Fin and then Fin 2, and was in 2025 discussions with investors about multibillion‑dollar valuations alongside other AI customer‑support vendors like Kore.ai. These companies are positioned as differentiated platforms built on top of foundation models, not as commodity utilities displaced by them. (en.wikipedia.org)
3. Market research describes a fast‑growing, multi‑vendor software market, not a commodity layer owned by a few model providers. Industry reports put the AI‑for‑customer‑service market at roughly $12–13B in 2024, projecting high‑20s CAGRs and tens of billions in value by 2030, with long lists of distinct software vendors (Zendesk, ServiceNow, Freshworks, Ada, Aisera, Intercom, LivePerson, etc.) participating in the space. This is consistent with a competitive applications market riding on foundation models, not one that has been “run over” and reduced to undifferentiated infrastructure. (globenewswire.com)
4. Commentary from practitioners and vendors frames AI features as ‘table stakes’ but stresses differentiation in implementation, trust, and integration. Multiple 2024–2025 pieces from customer‑service‑AI providers emphasize that simply having a chatbot is now “table stakes,” yet argue that real differentiation comes from agentic behavior, omnichannel workflows, governance, and domain‑specific tuning—i.e., application‑layer work on top of commoditizing models still matters and is where value is captured. (forbes.com)
5. Gartner/Reuters analysis points to a crowded and hyped vendor landscape, not pure model‑layer domination. Gartner estimated thousands of self‑described “agentic AI”/support vendors, with only ~130 considered technically legitimate, and predicted that many projects will be scrapped. That suggests oversupply and shake‑out risk—but also that a subset of vendors have meaningful, defensible technology and business value, rather than all value collapsing into generic foundation‑model APIs. (reuters.com)

Taken together, the data show a crowded, partially commoditizing feature space (everyone can plug into foundation models) but continuing strong value capture by differentiated AI customer‑service platforms and startups well past September 2025. That contradicts Chamath’s strong claim that you “cannot touch customer service” because it would be fully commoditized and “run over” by foundation models within about a year.