Owners ask one question more than any other on the first call. "If we said yes today, when could we have something working?" The answer matters, because in most cases the value of an AI agent is the value of the time it gives back, and every week of build is a week the team is still doing the work by hand.
Here is a realistic look at timelines for AI agent projects in 2026, with what changes the number.
The honest baseline
For a focused first project, three to five weeks from kickoff to live is normal. Anything quicker than that is usually either a SaaS chatbot in disguise or a project that has not been properly scoped. Anything much longer than eight weeks is usually a sign the scope has crept or there is something genuinely complex happening.
Inside that, three rough categories.
Two to three weeks: small and tight
Projects that ship in two to three weeks are tightly scoped, talk to one or two systems, and have a clear definition of done.
- An internal knowledge agent over a defined corpus of company documents, deployed on Slack.
- A first-response sales agent that handles inbound enquiries, qualifies, and books a call.
- A document drafting agent for one specific document type.
- A product description generator for an ecommerce store, set up against your tone of voice.
These are achievable in two or three weeks because the integrations are minimal and the success criteria are unambiguous.
Four to six weeks: the typical first build
Most first builds for UK SMEs sit here. Multiple integrations, a workflow with three or four steps, sensible escalation to a human, real testing before going live.
- A customer support agent that pulls live order data, handles returns, and escalates cleanly.
- A back-office agent that processes invoices, matches purchase orders, and writes to your accounting system.
- A maintenance triage agent for a property firm.
- A booking and reminder agent for a clinic.
This is also where most projects feel right. Long enough to do the job properly. Short enough to keep momentum.
Six to ten weeks: more involved
Projects in this range usually have one or more of: complex integrations, voice components, multi-agent workflows, or sectors with serious compliance requirements.
- A voice agent for inbound calls out of hours, with proper handoff to a human.
- An end-to-end claims handling agent for a logistics business.
- A clinical-adjacent agent for a healthcare provider with proper data residency and audit logging.
- An agent that orchestrates several smaller agents to do a multi-step process.
What actually drives the timeline
Five things change the number more than anything else.
- Scope discipline. A clear, narrow scope ships fast. A broad, "while we are at it" scope drags on.
- Integrations. Each system the agent talks to is real engineering work. Modern APIs are quick. Legacy systems with file-based exports are not.
- Decision-makers being available. If sign-off takes a week, the project takes longer. Builds need a single, available decision-maker on the client side.
- Test data quality. If you cannot give the agent realistic examples to work against, it cannot be tested properly. Late-stage testing is the most common reason projects drag.
- Compliance work. DPIAs, security reviews, internal sign-off processes. All necessary, all add real time.
The fastest projects we ship are the ones where the client knows exactly what success looks like, can give us real examples, and has one person empowered to say yes.
Why projects drag
The two most common reasons builds run late.
First, scope creep. "While we are at it, can it also..." kills first projects. The right answer is "yes, in the next phase". Add things, do not absorb them.
Second, the client side underestimates the time needed for review. Most owners assume they can squeeze sign-off into half an hour. In practice, looking at outputs properly takes time, and a rushed review usually means a missed issue that costs a week to fix later.
What we ask of clients to keep things on time
- One named decision-maker, available within a working day.
- Real examples of the work the agent will be doing, drawn from the last few months.
- Honest sign-off slots in the calendar, not squeezed in at the end of a Friday.
- A small group of pilot users on the team to actually use it for the first two weeks.
None of this is heavy. All of it makes the difference between a project that lands on time and one that drifts into "where are we with this".
The short answer
For most UK SMEs, plan on four to six weeks for the first AI agent. Anything quicker is suspicious. Anything much slower has usually been mis-scoped.
If you would like an honest estimate against your specific use case, tell us roughly what you are thinking and we will come back with a real timeline, not a sales pitch. For the cost side of the same question, our piece on how much a custom AI agent costs in 2026 is a good companion read.