How long does it take to build an AI agent?
Short answer
A first AI agent that handles one workflow takes 2 to 8 weeks to build in 2026, on timelines published by agencies that sell the work. SpotDev, a UK agency, says a first rollout is typically live in two to three weeks, and RaftLabs puts a single-workflow agent at 4 to 8 weeks. A production agent connected to three to five systems takes 12 to 20 weeks. Access to your systems, messy data and approvals set the pace more than the AI does.
Deniz SancarFounder of Yarigai, London. Updated .
Yarigai is a done-for-you AI agency that builds these systems, so read this with that in mind. Every price and figure links to its source below. Read this page as Markdown.
At a glance
| Question | Answer |
|---|---|
| First agent in production, UK agency | 2 to 3 weeks (SpotDev, May 2026) |
| Single-workflow agent or proof of concept | 4 to 8 weeks (RaftLabs, December 2025) |
| Simple custom chatbot, one channel, no CRM | 4 to 6 weeks (Softomate, updated October 2026) |
| Production agent with 3 to 5 integrations | 12 to 20 weeks (RaftLabs) |
| Multi-agent or enterprise system | 16 to 30 weeks (RaftLabs) |
| Connecting a CRM with a documented API | 2 to 3 days (Softomate) |
| Legacy system with no API | Adds 2 to 3 weeks (Softomate) |
| Share of delays caused by slow data from the client | 80% on Softomate's own projects |
| Agentic AI projects Gartner expects to be cancelled | Over 40% by the end of 2027 (June 2025) |
Every timeline here comes from an agency that sells AI builds. This guide found no independent survey of AI build times.
How long each type of AI agent takes
Published timelines run from 2 weeks for a narrow first agent to 30 weeks for an enterprise multi-agent system. Scope and the number of connected systems explain most of the gap.
| Build type | Published timeline | Source and date |
|---|---|---|
| First agent in production | 2 to 3 weeks | SpotDev, 27 May 2026 |
| Website FAQ bot with documents ready on day one | 3 weeks minimum | Softomate, updated 1 October 2026 |
| Simple chatbot (one channel, under 50 documents, no CRM) | 4 to 6 weeks | Softomate |
| Standard chatbot (two channels, one CRM with read and write) | 6 to 8 weeks | Softomate |
| Complex chatbot (three or more channels, two or more systems, compliance audit) | 8 to 16 weeks | Softomate |
| Single-workflow agent or proof of concept | 4 to 8 weeks | RaftLabs, 24 December 2025 |
| Production agent (3 to 5 integrations, memory, monitoring) | 12 to 20 weeks | RaftLabs |
| Multi-agent system or enterprise build | 16 to 30 weeks | RaftLabs |
| Workflow or scheduling agent | 8 to 12 weeks | Groovy Web, updated 23 October 2026 |
| Document processing and extraction | 8 to 14 weeks | Groovy Web |
| Voice or transcription product | 6 to 10 weeks | Groovy Web |
RaftLabs puts discovery at 2 to 4 weeks and evaluation and testing at 2 to 4 weeks. Softomate says adding voice escalation to a chatbot adds 3 to 5 weeks, and a WhatsApp project rarely goes live in under 5 weeks from contract signature.
What takes the time in an AI agent build
The model is rarely the slow part. Access to systems, data clean-up, testing on real past work, approvals and staff training take most of the calendar.
Access to your systems
Softomate says 80% of delays on its chatbot projects come from slow data provision by the client, with late system access and unclear escalation rules adding more. It estimates 2 to 3 days to connect HubSpot, 2 to 4 days for Zoho CRM, 3 to 5 days for Salesforce and 5 to 10 days for a bespoke CRM or ERP with a REST API. A legacy system with no API adds 2 to 3 weeks. RaftLabs says an undocumented API or on-premise access makes a timeline grow fast.
Data clean-up
Groovy Web says data that is unclean, locked in legacy systems or stuck in PDFs adds 2 to 6 weeks. SpotDev says that when data is scattered, duplicated or incomplete, the groundwork "has to happen first".
Testing on real past work
RaftLabs budgets 2 to 4 weeks for evaluation and testing, and Groovy Web plans one or two evaluation cycles of 1 to 2 weeks each. The useful test runs the agent on a batch of past work where you already know the right answer, such as last month's support emails or invoices, and compares its output with what your team did.
Approvals and compliance
Groovy Web says healthcare, finance, legal and HR projects add 4 to 8 weeks for compliance. Softomate says a data protection officer or legal contact named late can push go-live from weeks 6 to 8 out to weeks 7 to 9, and FCA, CQC or SRA obligations add compliance work. At enterprise level, RaftLabs says security reviews and legal sign-off on vendor contracts routinely add 4 to 8 weeks.
Staff training and the first month live
Softomate runs a soft launch of 5 to 10 working days and 30 days of close support after full go-live. Staff need to know what the agent does, where it stops for approval and how to flag a wrong answer. The people who approve its work need time in their week to do it.
A week-by-week example plan
This example covers an agent that reads incoming customer emails, drafts replies from your help documents and logs each one in your CRM, with a person approving every reply. It is an illustration built from the published phase lengths above, not a quote.
- Week 1. Map the workflow, agree what counts as a correct reply, collect past emails with the replies your team sent, and grant access to the inbox, help documents and CRM.
- Week 2. Connect the inbox and CRM, which takes 2 to 3 days for a CRM with a documented API on Softomate's estimates, and load the help documents.
- Week 3. Build the first version and run it on the past emails. Log every wrong or weak draft.
- Week 4. Fix the failures, add the approval queue and set hand-off rules for complaints, refunds and anything the documents don't cover. Run the past-email test again.
- Week 5. Run in shadow mode. The agent drafts, staff reply as normal, and you compare the two. This matches Softomate's soft launch of 5 to 10 working days.
- Week 6. Go live with approvals switched on, train the team, and read failed runs every day for the first 30 days.
Six weeks sits inside RaftLabs' 4 to 8 weeks for a single-workflow agent and Softomate's 6 to 8 weeks for a standard chatbot with one CRM connection. A legacy system with no API would add 2 to 3 weeks, and a regulated firm should add compliance time on top.
What speeds up an AI agent build
The fastest builds have one workflow, access granted in week one and a person on the client side who answers within a day.
- Pick one workflow with one measure, such as hours saved a week or emails answered without a person.
- Grant access to every system in the first week. Softomate says each week of client silence during testing adds a week to delivery.
- Have past examples and documents ready. Softomate's knowledge base phase shrinks to 5 working days for a 30-page FAQ.
- Name the person who signs off data protection at the start.
- Agree the pass mark before testing begins, so testing has an end point.
- Check whether you need an agent at all. Anthropic recommends "finding the simplest solution possible", and a fixed list of steps is an automation that is quicker to build. The guide to AI agent or automation sets out the difference.
Why AI agent projects stall
Most stalled projects lack a clear business case and real testing, not working code. Gartner predicted on 25 June 2025 that over 40% of agentic AI projects will be cancelled by the end of 2027, because of rising costs, unclear business value or inadequate risk controls. Gartner analyst Anushree Verma said many of these projects are "early-stage experiments or proof of concepts that are mostly driven by hype". In a January 2025 Gartner poll of 3,412 webinar attendees, 19% of organisations had made significant agentic AI investments and 42% had made conservative ones. Gartner also estimated that only about 130 of the thousands of vendors selling agentic AI offer real agentic features.
MIT's Project NANDA reported in July 2025, in a report summarised by Virtualization Review, that 95% of organisations saw no business return from generative AI, despite $30 billion to $40 billion of enterprise spending. It blamed brittle workflows, weak contextual learning and misalignment with day-to-day operations, and found that only 5% of custom enterprise AI tools reach production.
The stall points show up in the same places on small projects:
- The scope grows mid-build. Groovy Web names scope creep and model selection paralysis as the most common reasons projects run late.
- Nobody owns the business case, so nobody can say whether the agent is worth switching on.
- The agent never runs on real past work, so staff don't trust it.
- Data protection and approvals are left until the end, when they add weeks.
How Yarigai plans an AI agent build
Yarigai is a done-for-you AI agency in London that builds AI agents inside the client's own accounts and tools. Its timeline is set in four steps.
- A free 30-minute scope call on Google Meet. The client keeps the notes whether they hire Yarigai or not.
- A blueprint, which takes about a week and is credited to the build. It sets out a written plan, a clickable prototype, the monthly running cost, a fixed price and the payback maths.
- The build, which takes 2 to 6 weeks in the client's accounts, with a demo of working software every week. A single workflow can go live in about two weeks.
- Optional monthly care, which the client can cancel any month and keep everything.
By that arithmetic, a project runs about 3 to 7 weeks from the start of the blueprint to go-live. The weekly demo shows progress on working software, so a delay is visible in the week it happens. The client owns the code, the data and the accounts from day one, and a person approves anything that sends money, emails a customer or deletes data.
Yarigai quotes a fixed price after the one-week blueprint. Its guarantee reads "If the build isn't profitable for you, we won't build it." If the blueprint maths doesn't come out ahead within the first year, Yarigai doesn't take the job. More on AI agents and the guarantee.
Common questions
Can an AI agent be built in a week?
A prototype can, but a production agent rarely is. Groovy Web says a prototype can be built in under 4 weeks but needs 3 to 6 more weeks of engineering before it is ready for production. The shortest published UK figure for a first agent in production is SpotDev's two to three weeks.
How long does a simple AI chatbot take to build?
Softomate puts a simple custom chatbot, with one website channel, under 50 documents and no CRM, at 4 to 6 weeks. Its minimum is 3 weeks when the documents are ready from day one. Adding voice escalation adds 3 to 5 weeks.
Why do AI agent quotes give such different timelines?
Mostly because of integrations and scope. RaftLabs says timelines are "dominated by integration engineering", and an undocumented API adds time fast. Ask each supplier which systems they have counted and whether testing is inside the quoted weeks.
How long does it take to connect an AI agent to a CRM?
Softomate estimates 2 to 3 days for HubSpot, 2 to 4 days for Zoho CRM, 3 to 5 days for Salesforce and 5 to 10 days for a bespoke CRM or ERP with a REST API. A legacy system with no API adds 2 to 3 weeks.
How long does testing an AI agent take?
RaftLabs budgets 2 to 4 weeks for evaluation and testing, and Softomate runs a 5 to 10 working-day soft launch before full go-live. Testing ends sooner when you agree a pass mark first and test on your own past work.
Does an AI agent take longer to build in a regulated industry?
Yes. Groovy Web says healthcare, finance, legal and HR projects add 4 to 8 weeks for compliance, and Softomate says FCA, CQC and SRA obligations add compliance work. Name your data protection lead at the start to keep that time down.
Is an in-house team faster than an agency?
Not for a first project, on the evidence available. Groovy Web says a first-time AI team adds a 30% to 50% learning curve and a mixed team that splits its time runs 1.4 to 1.8 times longer. Those figures come from an agency that sells builds, so weigh them as a vendor view.
Sources
Checked on or before 7 November 2026. Prices change, so check the live page before you decide.
- How Much Do AI Agents Cost in the UK? A Transparent Breakdown, SpotDev, published 27 May 2026, checked 7 November 2026.
- AI Agent Development Cost in 2026: What You'll Actually Pay, RaftLabs, published 24 December 2025, checked 7 November 2026.
- How Long Does It Take to Build a Custom AI Chatbot? A Real 4-8 Week UK Timeline with Milestones, Softomate Solutions, updated 1 October 2026, checked 7 November 2026.
- How Long Does It Take to Build an AI Product? Real Timelines for 2026, Groovy Web, updated 23 October 2026, checked 7 November 2026.
- Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027, Gartner, published 25 June 2025.
- Gartner: More than 40% of agentic AI projects will fail by 2027, RCR Wireless News, published 27 June 2025, checked 7 November 2026.
- MIT Report Finds Most AI Business Investments Fail, Reveals 'GenAI Divide', Virtualization Review, published 19 August 2025, checked 7 November 2026.
- Building effective agents, Anthropic, published 19 December 2024, checked 7 November 2026.
Keep reading
How much does it cost to build an AI agent for a small business?
A first AI agent costs £6,500 to £8,000 to build on published UK agency price lists in 2026, plus about $6 to $240 a month in model usage at 2,000 runs.
Read moreAI agent or automation: what's the difference and which do you need?
An automation runs fixed steps and costs from £1,500 to build in the UK in 2026. An AI agent chooses its own steps and starts from £6,500. Here is how to choose.
Read moreWho can build an AI agent for my business?
Five routes to an AI agent in 2026, from a £127 Fiverr gig to an £8,000 agency build or an £87,500 hire, compared on cost, ownership and who fixes it.
Read moreWhere should a small business start with AI in 2026?
Start with one costed workflow. DSIT says 32% of UK firms used AI in 2026 and 65% of non-users see no clear business case. How to pick and measure it.
Read moreWant a dated plan for your agent?
Book a free 30-minute scope call and keep the notes. If the build isn't profitable for you, we won't build it.