What is Tools-as-a-Service and where does it fit?

Tools-as-a-Service is a new category. This piece defines it: what it is, what it does, and where it sits in the agent stack relative to the categories that adjoin it.

If you're building AI agents on top of SaaS products, this is the layer you've probably been building manually — or not building at all, and hitting the consequences.

The agent stack, briefly

AI agents have four core components:

Model — the reasoning layer. Takes in context and produces decisions about what to do.

Orchestration — manages the loop between reasoning and acting. Handles memory, multi-step planning, tool selection.

Tools — define what actions the agent can take. When the agent decides to do something, it invokes a tool to do it.

Memory — provides context. What has happened in this session, what the user's state is, what the agent has already done.

Tools-as-a-Service operates in the Tools layer.

What Tools-as-a-Service is

Tools-as-a-Service (TaaS) is a platform that generates, maintains, and runs the tools that let agents act on SaaS products.

The three components:

Tool generation. Produces structured, typed tool definitions from a SaaS product's existing API surface and codebase. The generation layer scans the codebase and produces two artefacts: connectors (atomic action definitions, mapped to specific API capabilities) and tools (higher-level workflow definitions composed from connectors, representing business tasks a user would recognise). This is not code generation for SDK clients — it's the production of agent-callable tool definitions from a product's real capabilities.

Tool maintenance. Keeps tool definitions current as the underlying product changes. When an API changes, the tool definitions update — aligned with the product's own SDLC rather than discovered when an agent invocation fails in production. Automated maintenance is what makes tool generation a platform rather than a one-time output.

Tool runtime. Executes tool calls reliably and securely. Handles auth lifecycle management (obtaining, refreshing, and revoking credentials), rate limiting, failure handling, and observability. Tool calls execute as the authenticated user — with the user's permissions, data visibility, and audit trail — not as a service account.

Together, these three components provide the tools layer of the agent stack as a managed service: something an agent team can build on without building the tool generation, maintenance, and runtime infrastructure themselves.

What Tools-as-a-Service is not

Four adjacent categories it's useful to distinguish from:

Agent frameworks (LangChain, OpenAI Agents SDK, CrewAI) handle orchestration — the loop between reasoning and acting, memory, multi-step planning, tool selection. TaaS provides the tools; agent frameworks decide when to invoke them. Complementary, not competing.

SDK generators (Stainless, Fern, Speakeasy) generate client libraries from OpenAPI specs. TaaS generates tool definitions from codebases, including capabilities the API spec doesn't expose. Different input, different output, different purpose.

Integration platforms (MuleSoft, Boomi) move data between systems — typically in both directions, for a wide range of integration patterns. TaaS generates tools from a SaaS product's codebase to expose that product to agents. Different direction, different use case.

MCP gateways provide governance between AI clients and MCP servers — RBAC, DLP, observability on top of existing tool definitions. TaaS generates and maintains the tool definitions themselves. Different layer of the stack.

Who Tools-as-a-Service is for

The customer for a TaaS platform is a B2B SaaS company that:

• Is building AI agents that need to act on their own products
• Has a gap between their API coverage and the capabilities their agents need
• Can't close that gap through a full API rebuild on a viable timeline
• Has a product portfolio complex enough that bespoke connector work doesn't compound

The end users of a TaaS platform are the agents themselves, and the users those agents act on behalf of. A TaaS platform exposes an Account's product capabilities to agents operating as the Account's customers.

Why this layer exists

The tools layer has to exist. Agents need tools to act. The question is who builds and maintains them.

Foundation model providers — OpenAI, Anthropic, Google — are building tool ecosystems for the top 20% of SaaS products by integration volume: the Salesforces, the Slacks, the HubSpots. The long tail of vertical SaaS, industry-specific tools, and portfolio products acquired through M&A won't appear on a foundation model provider's tool integration roadmap.

For those products, the tools have to come from somewhere else. TaaS is the answer — a platform purpose-built to generate, maintain, and run agent tools for the products that won't get covered by the horizontal platforms.