AI agents for business: where they fit, and the controls they need

The four terms are used loosely, vendors included. These are the working definitions this guide uses, each tied to a published source listed at the end of the page.

Agents earn their cost on work that needs judgement across inputs no rule set captures, and where a wrong step can be caught before it lands. OpenAI’s guide names three signs: decisions that need judgement and are full of exceptions, rule sets that have grown too costly to maintain, and heavy reliance on unstructured material such as documents and conversations. Where a use case does not clearly meet those tests, the same guide says a deterministic solution may be enough.

The table below is our reading of how that plays out across common business functions. It describes where to look, not results anyone has measured in your business.

One question comes before all of them: does the model need to choose the steps at all? Anthropic’s published advice is to find the simplest solution possible and add complexity only when it is needed, and it notes that agentic systems often trade speed and cost for better task performance. Many useful systems are a fixed workflow with one model step inside it, which is cheaper to run and easier to test than an agent.

Three kinds of agent now reach a business, and each is bought and checked in a different way. Coding agents such as Claude Code and Codex work in a repository, where tests and code review already stand between a change and production. Office agents such as ChatGPT Work and Claude Cowork work on documents, email and the apps a person connects, and they come inside the vendor’s plans rather than being sold alone. A workflow agent built for one process, the subject of the rest of this guide, runs against your own systems of record.

I would buy these per team rather than per vendor, and set one permission and review model above all of them. The table shows where each kind fits and what already checks its work.

Codex and ChatGPT Work share one usage allowance, and Claude Cowork draws on the same Claude plan limits as chat and Claude Code, so a person who uses both kinds of agent from the same vendor draws on one allowance, not two (on Claude Enterprise, usage is billed at API rates). Our side-by-side of Claude Code and Codex for engineering teams covers the coding half, and our guide to how ChatGPT Work and Claude Cowork differ covers where each office agent runs and what admins control.

Four situations argue against an agent. When the steps are known and stable, a workflow or a scheduled job does the same work more cheaply and more predictably. When nobody can say what a correct result looks like, nothing can check the agent either. When every output is read by a person before anything happens, a drafting assistant is enough, because the reader is the check. And when an error would land before anyone could see it, the agent should not be allowed to act until the controls in the next section exist.

Multi-agent systems add a further cost. Anthropic reports that, in its data, agents use about four times the tokens of a chat interaction and multi-agent systems about fifteen times, and that work where every agent needs the same context, or where the agents depend heavily on one another, is not a good fit for multi-agent systems today. OpenAI’s guide likewise recommends getting as far as possible with a single agent before adding more.

Accuracy is the wrong measure for an agent. What matters is what happens on the occasions it is wrong, and whether anything notices before the consequence lands. The controls below are the ones to have in place before an agent may write to anything that matters. The agent proposes and deterministic checks decide what proceeds; anything that fails a check goes to a person with the reason attached. The mechanism is set out in full in our published pattern for gating an agent’s actions.

Together, these controls are what turn a pilot into an enterprise AI agent. The model is often the same one the pilot used. What changes is that each control below has an owner and leaves evidence before the agent acts on a live system.

Vendors now ship parts of this in their agent products. The OpenAI Agents SDK includes guardrails that validate an agent’s inputs and outputs, and built-in tracing of each run. Anthropic’s Claude Agent SDK lets a developer set which tools run automatically and which need approval, and run custom code at set points in the agent’s lifecycle. Microsoft’s Agent Framework supports tools that pause a workflow for human review before they run. These are building blocks. Which actions need a check, and who approves them, is still the business’s decision. OpenAI’s own guide names two triggers for handing control to a person: an agent exceeding its failure limits, and actions that are sensitive, irreversible or high-stakes.

Two published engagement files describe parts of this. At a large international marketing agency, we build managed-agent workflows for finance-critical tasks, so deterministic checks decide what proceeds (engagement file D-01). In engagement file D-06, a connector answers questions from finance data while carrying each person’s existing permissions, so nobody sees more than they already could.

Where an agent handles personal data, data protection law applies to it as to any other processing. The ICO’s January 2026 report on agentic AI says organisations remain responsible for the data protection compliance of the agentic AI they develop, deploy or integrate, and the OWASP GenAI Security Project published a Top 10 for Agentic Applications in December 2025 setting out the main security risks. This is not legal advice. Which rules apply to a particular agent is a question for your legal adviser or data protection officer. For where these controls sit in a wider policy, our governance guide shows an AI policy mapped to its controls, clause by clause.

The figures in this section are very high level and indicative. The true cost of a project can only be established after a discovery stage, because work such as sorting out data and infrastructure often has to come before an agent is built, and nobody can price that work without looking at it. Use the bands to sense-check a quote, not to set a budget.

They are our reading of agency price guides published in 2026, read on 30 September 2026, and they are in US dollars because the guides mostly publish in dollars. For each tier we took the product line in each guide closest to that scope and applied one rule: the typical band runs from the median of the published floors to the median of the published ceilings. The lowest floor is the cheapest start among the product lines we mapped to that tier: Creole Studios (September 2026) lists AI-powered chatbots from $10,000 and custom workflow agents from $25,000, and Azilen (February 2026) lists multi-agent systems from $150,000. These are sellers’ published ranges, not what buyers paid, and none of them is a 1AYM price.

The awards sit in their own column because they measure something else. They are real transactions, but the amount a US agency obligates can include hosting, support, security accreditation and option years, and some awards run for several years: the Peace Corps chatbot award is $75,750 now and $166,560 with all its options, to September 2031. They show what public programmes spend, not the price of one agent, so they are never blended into the typical band.

The model is rarely what sets the price. Sharp Code (August 2026) says the software, data and change work around the model drive the bill, and adds 20 to 30 percent for a legacy ERP or CRM. Azilen (February 2026) puts integration engineering and testing at 40 to 60 percent of the build. That is the same story as the drivers below, and it is why a discovery stage comes first: it finds the data and infrastructure work before that work turns up halfway through a build.

Building the agent is one line of a larger budget. Our breakdown of AI implementation cost, line by line, covers discovery, integration, change management and running costs around it. A price without a scope is a guess, so the reliable route to a number is to name one workflow, count the systems it reads and writes, find out what state their data is in, and say what a wrong answer costs. 1AYM will tailor the project to your budget and timeframe, and your account manager will make sure expectations are set from the first conversation. I would read any quote far outside these bands as a prompt to ask what it includes, or what it leaves out.

No two agent projects cost the same, and a single published price would hide the drivers below. Each one can be estimated before a build starts.

Once an agent is live, its tokens and running time become a monthly line with an owner, and our guide to budgeting and capping AI spend covers the limits that hold a running agent to a number.

The runtime is increasingly a vendor product rather than something to build. The vendor runtimes we call an AI harness already include an agent loop, tools and permission settings: Anthropic’s Claude Agent SDK, for example, packages the loop, tools and permissions that run Claude Code. What a business still owns is the work around the runtime: the definitions, the checks, the access model and the person who answers when it fails.

Coding agents are the same pattern pointed at a repository, and our comparison of Claude Code and GitHub Copilot shows where two of them keep their permission settings and who can approve their work.

Use this before you commission a build, from us or from anyone else. Each item should have a written answer before any code is written.

If the checklist shows the workflow is not ready, that is a useful answer, and cheaper to learn now than after a build. If you want help with it, a fixed-scope AI strategy sprint of typically two to four weeks scores candidate use cases on value, feasibility and risk.