By Bryan Knipe, founder of TriPinnacle Advisors
Most companies use AI to make individual tasks faster: a summary here, a first draft there. The function stays the same size, with the same people doing the same jobs a little quicker. That saves something, but not much, and it rarely shows up in the P&L.
The bigger change is running whole functions on AI. Research, reporting, finance operations, outreach and follow-up, document review and the first draft of nearly everything can be done by AI agents that I build and run, with people kept where judgment, relationships and accountability live. Done well, a company does the same work, or more, with far fewer people.
The person doing this has to know how the function should run, not just how the technology works. I have run these functions myself, as a COO/CFO and general manager, a sales leader, and the head of analytics and reporting organizations, after ten years as a software engineer and architect. That is what lets me redesign a function around AI instead of automating the old version of it.
I run my own software company this way. As one person, I cover product, go-to-market, operations and finance, with AI agents doing the work a team would otherwise do: AI coding agents build the product to my design, a research agent combs public sources for investors, grants and partners, and agents draft materials, outreach and follow-ups for me to review. I have also done it for clients: at one startup, moving manual work onto AI tools I built was a large part of taking the monthly burn from about $150,000 to about $40,000. This is how I approach it.
1. Map the company, not the tasks
Start with the functions: what each one produces, for whom, how often, with how many people, and what it costs. For each, I want to know how much of the work is reading, writing, looking things up, moving data between systems, checking and reporting. That share is the part AI can take. In most finance, operations, sales support and research functions, it is most of the work.
2. Pick functions by volume and rules
The first functions to move are the ones with high volume, clear rules for a good result, and work that people check by hand today: reporting, reconciliations, invoice and contract handling, research and list-building, outreach and follow-up, first drafts of standard documents. Leave the work that is rare, high-judgment or relationship-driven with people. Move one function at a time, so each one is proven before the next.
3. Clean up before you automate
Automating a messy function makes the mess faster. Before building anything, cut the reports nobody reads, the steps that exist because of an old mistake and the tools that duplicate each other. I cover this in the guide on cutting costs with AI; it is usually where the first savings come from.
4. Build agents that do the work, not chatbots that answer questions
An AI agent that runs a function takes inputs, does the multi-step work, and produces the output the function owes the business: the report, the reconciled ledger, the researched list, the drafted follow-up. The ones I build run under an orchestrator that logs every step, meters every model call and stops at a hard cost limit, so a bad run is visible and bounded, and cost per unit of work is a known number.
5. Keep people on judgment, relationships and accountability
The design question is not "what can AI do?" but "where must a person decide?" In my own company, the agents propose and I decide: nothing goes to an investor, a customer or the company record without a person accepting it. The same rule works in any company. The people who remain do the judgment work the function always needed and never had time for, which is why the smaller team is usually a better-paid and more capable one.
6. Check the agents' work
Every output an agent produces is checked against its sources before anyone acts on it, and answers that cannot be supported are dropped rather than shown. People handle the exceptions the system flags. As the error rate proves itself on real work, the share needing a person shrinks.
7. Measure it against a baseline
Before a function moves, measure what it costs and how it performs. Afterwards, measure again, including the AI cost, the reviewers and the time spent on errors. For an owner or a board, that comparison is the whole case. For a private-equity owner, it is margin that holds, because the work is done differently rather than just done by fewer people.
8. Handle the people side honestly
Moving functions onto AI changes jobs. Decide early what happens to the time it frees: redeployment to work that was neglected, growth, or a smaller team. Say which, and treat people well. The people doing the work today are the best source of what to automate, and they will help when they trust how the answer will be used.
Where I help
I do this as a forward-deployed operator for founders building lean from the start and for private-equity owners who want portfolio companies to do more with less. I don't need a detailed brief: I map the company, decide what moves first, build the automation and run it in the work until it holds, part-time alongside your team or as an interim COO.