The consensus from this morning's intelligence briefings lands on a single, shared thesis: the CRM is not dying, it is being demoted to plumbing. Meanwhile, AI agents are becoming the new surface where work actually gets done. Both a16z and NFX arrive at the same conclusion from distinct vantage points.
Andreessen Horowitz's latest deep-dive draws a sharp historical parallel: what Facebook's News Feed did to the friend graph, AI agents are now doing to Salesforce. The database once was the product. Now it is "just one of many inputs" fed into the reasoning layer above it.
Pete Flint at NFX makes the same argument in a tighter frame, deploying the Kodak Brownie as his central metaphor. When Kodak released a $1 camera, it did not destroy the photography industry — it created ten million photographers. Technology expands markets. It does not shrink them. The labor budget grows alongside the software budget, because the returns per dollar of GTM spending rise with AI-augmented reps.
For twenty years, owning the System of Record was the winning play in go-to-market software. Salesforce and HubSpot sit on some of the most valuable datasets in the industry. But a16z argues that the locus of value is migrating upward — into the layer that reads and writes to the database and does the actual thinking.
The new stack: foundation models at the bottom, domain-specific orchestration in the middle, and the user experience on top. Between the model and the customer sits enormous unglamorous work: orchestrating context across dozens of systems, encoding sales logic, handling permissions and compliance, integrating with Fortune 500 IT environments.
A few years hence: she begins her day not in a static Salesforce pipeline view, but in a prioritized feed generated by her system of intelligence. Which accounts had material news overnight? Which prospects are suddenly in-market? Which deals have gone quiet and need investigation? The daily prioritization decision — once consuming real cognitive effort from every rep — has been quietly offloaded.
Switching costs migrate: "All our customer data is in Salesforce" becomes "all our workflows, reasoning, and accumulated institutional context live in our AI layer." The CRM remains indispensable — but as a database consumed at the API layer, not as the surface where work happens.
The fear is that AI destroys jobs, industries, and opportunity. Flint calls this "dead wrong." Technology consistently creates more than it destroys — but it creates in different places, and those who position themselves at the new layer capture the value.
The category formed because model supply fragmented faster than buyers could adapt. Over 100 frontier-grade models shipped in twelve months. Each demanded a new SDK, auth scheme, rate limit, and contract review. The cost of moving exceeded the cost of being slightly wrong.
If changing one parameter switches vendors, the product is an AI Gateway. If the user has to write a Dockerfile, it is AI Infrastructure. These two markets compete on entirely different surfaces.
| Player | Play |
|---|---|
| OpenRouter | 300+ models, one endpoint. The reference open aggregator. |
| Cloudflare AI Gateway | Edge-distributed, caching, unified billing, analytics. |
| Vercel AI Gateway | Zero markup on tokens, $5/mo free tier, tight AI SDK integration. |
| Portkey | Enterprise control plane. Acquired by Palo Alto Networks (Apr 2026). |
| Kong AI Gateway | Throughput-oriented, extends Kong's API management into AI traffic. |
| Fal.ai | Generative media gateway — image, video, audio, 3D behind one endpoint. |
One. Token markup collapses to ≤5% across every major gateway. Vercel already prices at zero. Once one well-funded player commits to passthrough, peers follow.
Two. Margin moves upstream — to caching, observability, evals, governance, and integration depth. The pure-aggregation layer has no moat.
Three. At least one frontier lab ships a multi-provider gateway routing to competitors. The lab that does this earns trust to be the default routing layer when its own model is the right answer.
Four. Gateways become a default feature inside every major cloud platform. AWS Bedrock and Azure AI Foundry are already converging on the same shape.