Why we publish our engines: building for AI agents, now

articleAscent Partners12/10/2026

Most software is written for a person to open, read and click. A growing share of work is now done by AI agents — software that reads, decides and runs the next step on its own. We think that shift is not a side current; it is where knowledge work is heading.

For a decision-support firm, that matters in a specific way. If an agent is going to act on a valuation, a disclosure or a control assessment, it needs the same things a human reviewer needs: the method, the standard it cites, and the working behind the number. A plausible answer is not enough. It has to be defensible.

So we took a position early. Rather than wait for the market to settle around agents and then adapt, we build the engines for them now — as MCP servers and public APIs — and open them. The tools are the machinery behind our own numbers, published so an agent can use them directly.

What we publish

Our open tooling covers valuation, sustainability and corporate service. Each server is open — no API key — and returns results cited to the standards a reviewer will check: IVS, IFRS, ISSB, and for internal control, COSO. The most recent addition is the COSO Internal Control MCP, which runs an internal-control review end to end, from scoping and risk to sampling, testing, workpapers and deficiency evaluation.

Why open

An agent is only as trustworthy as the sources it can reach. If the engine is closed, an agent can only guess at it — or repeat a competitor's summary. Opening ours puts the method and the standards where an agent can use them, and where a human can inspect them. That is the same reason we cite standards at all: a decision that cannot be checked is not one you can defend.

Working with agents now also makes us better at it. Each tool is a small contract — deterministic, versioned, documented — and running them in the open teaches us what agents actually need. We would rather learn that with the tools in production than in a strategy deck.

The direction is clear enough. AI agents are becoming the default way work gets done. We are building for that mode now, and leaving the engines open for anyone — person or agent — to use.