Systems Plumbing Is Enterprise AI Bottleneck
Augusto Marietti, CEO of Kong, on the end of tokenmaxxing
The bottleneck for enterprise AI has shifted from model quality to systems plumbing. Agents are only useful when they can read from one system, act in another, and carry identity and permissions all the way through, but most large companies still run isolated apps, clouds, and data stores. That is why forward deployed engineering is back in force, because someone has to manually wire the rails before agent workflows can run reliably in production.
-
Kong is positioning its AI gateway as that rail layer. Its product sits in front of internal APIs and LLM traffic, then handles authentication, governance, routing, caching, and policy enforcement so enterprises can connect many models and systems without every team building its own glue code.
-
This is less a software only sale than a software plus implementation motion. The interview describes enterprises as still years away from true cross company agent workflows, and says straightforward engineering work is needed first. That is the same pattern seen in earlier enterprise products where hands on implementation teams drove adoption and retention.
-
The comparable is Palantir style deployment, not self serve SaaS. Airtable earlier used implementation specialists and integration engineers in a similar role, building each customer’s setup so the product fit real workflows. In AI, that work now centers on permissions, system mappings, and reliable tool calling across silos.
Over the next few years, the winners in enterprise AI will be the vendors that turn brittle internal systems into callable infrastructure. Once identity, policy, and routing sit in a common layer, forward deployed work can shrink from bespoke integration into repeatable deployment, and agentic software can start moving from narrow departmental copilots to cross functional execution.