A marketing director deciding which agency should build the team's first AI agent is buying something unusually hard to inspect. The work is invisible once it runs, the vocabulary is new enough that two agencies can describe different services in identical words, and the proof on offer is usually a percentage with no company attached to it. Of the eight UK agencies read for this list, only one shows working agents running inside three named businesses, with a written-up case study for each.
Disclosure: Agent Place shares an owner with The Keyword.
This list does not publish a score. A single number out of 100 tells a buyer where an agency finished and hides which test it failed, which is the part that matters when the failed test is the one that agency's own work depends on. Instead the eight are set against four checks, reported separately, with the first carrying the most weight.
The four checks
One, whether the agency runs this technology on its own business. This check carries the most weight here, because an ongoing agent service is not a project that finishes: the thing being rented is the agency's ability to keep software working after launch, and the only unpaid demonstration of that is its own operations.
Two, whether a client deployment can be opened and checked. A named organisation, a described result, and a page a reader can reach. A figure with the client withheld is a figure nobody outside the agency can confirm.
Three, what commercial terms are published. Price, the cost of keeping the thing running, and how an engagement ends. Agencies differ more on this than on anything technical.
Four, whether an individual is named. Not a team photograph or a shared inbox, but a person the published pages make accountable for the work.
The eight, in order
1. Agent Place
Agent Place is an AI automation agency and consultancy building bespoke agents for operations, sales and marketing, and it is the clearest answer in this group to the check that matters most: it runs its own agents inside working businesses, and publishes each one as a case study. An invoice-recording agent runs at the marketing agency TripleDouble. A weekly search-maintenance agent runs at this publication, whose own write-up of that agent sets out the method, the two figures that moved and what neither of them proves. A nightly product-documentation agent runs at the software company Wilow. A fourth case study shows three agents built for Hoxton Circle, a recruitment firm placing finance leaders.
For a marketing team, the search-maintenance agent is the most directly relevant piece of work in this list. It runs every week against a site's own Search Console data rather than a modelled estimate of it, and rewrites the pages losing clicks. The full range is on its agent pages, from invoice processing to product documentation.
The commercial terms are the simplest in the field: "Unlimited agents for one flat monthly fee", month to month, stoppable in any month, with nothing charged per call, minute or message, and nothing further paid until the first agent is working, inside 30 days. The agent pages are written and signed by Kole Ogundipe, so check four passes with a named person accountable for the work.
2. Agenticise
Agenticise publishes a specific claim about running agents on its own work, and it is about delivery rather than operations: "Every build runs on multi-agent orchestration. Discovery, planning, building, review and testing run as dedicated AI agents in parallel, with a person making the decisions."
Its terms are published more fully than most of this field. A focused audit starts "from around £3,000", a first build phase runs "£6,000 to £25,000", and the page puts maintenance at 10 to 20 per cent of the build cost a year — a cost most of this field leaves a buyer to discover later. Work is done "directly with David, our founder, throughout", which satisfies check four.
Check two is where it drops. Three results are published, each said to link to a case study, and every client is withheld behind a category: "Construction platform", "Global biometrics company", "Commercial flooring". One of the three is sales-operations work — 95 per cent less time spent qualifying leads — which is the kind of figure a marketing leader would most want to interrogate, and the one that cannot be attached to a company.
3. flowio
flowio publishes the most checkable client work in the group and no evidence at all of running agents on its own business. The Edinburgh Boiler Company is named, written up, and credited with more than £2m of sales growth and 533+ hours saved from speed-to-lead automation, CRM workflows and AI voice agents. Two further builds are written up anonymously, including a UK lettings agent with 70 per cent better lead qualification. For a marketing team, that is the closest thing in this field to a comparable engagement.
Its terms are the most completely published of the eight: packages "start from £750 for a single workflow automation", most clients "invest between £1,500-£15,000 for a complete automation solution", 90 days of support are included after launch, maintenance plans "start from £150/month", and it is the only agency of the eight publishing a refund term — a full refund if an automation does not deliver the agreed functionality within 30 days of launch. The founder, Malcolm, is named.
One caution, and it is on the numbers rather than the work: the site states its client savings as "£950K+" in its credentials strip and "£2M+" in the founder block. Two figures for one metric on one page is the sort of thing worth resolving on a first call.
4. Nectra
Nectra publishes one named, written-up deployment with the client's own director quoted — CK Fulfilment, a third-party logistics business, with an AI layer drafting customer replies inside its existing Freshdesk, more than 60 per cent of tickets auto-responded, replies 95 per cent faster, and every reply approved by a person before sending. The named quote from the client's director, Rebecca Kane, is one of only two client voices in the group; Ronins carries the other.
The costs are published at a single altitude: "most projects start at around £5,000". Nothing is published about running agents on Nectra's own business, and no individual at Nectra is named — the nearest thing is a promise about a type of person, that a reply will come "usually [from] a senior engineer rather than a sales rep". Its published approach is incremental: "We measure, then expand", with each further automation its own decision. A marketing team should note that the one deployment on show is customer-service work, not demand generation.
5. Ronins
Ronins is the most explicit agency in the group about the market it is not in: "we do not compete on £1,000 automations and we are honest about that from the first call". It publishes three ranges — discovery workshops at around £3,000 to £8,500, a proof of concept built on the client's own data at £12,000 to £30,000, and production systems with integration, governance and monitoring starting around £30,000 — and says it publishes them deliberately.
Three pieces of work are described in method and result with no client named in the page text, and two of the three are described as proofs of concept rather than finished deployments. The sales-relevant one is a CRM rebuild with three research agents over the top, flagging what each site buys compared with its sister sites. Ronins names its own agentic platform, SONAE, but that is reusable client-delivery infrastructure rather than evidence of agents inside its own business, and no individual at Ronins is named on the page.
6. Tiadara
Tiadara names the software it built and runs on its own site, and has the least client evidence behind any such claim in the group: "We created Prognosite, an SEO intelligence platform with 149 MCP tools. We use it to optimize this website", and the page goes on to offer the same build to a reader’s own site. Naming the thing is what makes the claim worth something — a reader can go and look at it, and for a marketing team an SEO platform is the most directly relevant piece of own-built software anyone in this field puts forward.
No client case study is published. On terms, the page gives a range — typical engagements of "£8,000 to £40,000" for a single end-to-end workflow, with ongoing cost described as model usage — but states no fixed price, and its scope is deliberately narrow: one process at a time, delivered in four to six weeks, which means a second agent is a second engagement. No individual is named.
7. Turing Automate
Turing Automate publishes an unusually plain exit position — the build and its documentation are the client's, the agreement can be paused or cancelled at any time, and the agency can be kept on to monitor it, fix it and add automations — and almost nothing else a buyer can check. No currency figure appears anywhere on the page. In place of client work it shows a lead-intake demo the page labels simulated, a live-looking activity log of invoice reminders, enquiry replies and a Monday report, and six of its own products. The only human name on the page, "Priya K.", is an entry in that log.
8. Molo Agency
Molo Agency comes last on these four checks while being, in one respect, the most straightforward: it states outright that it has published no client AI agent deployment, and its named, openable builds are deliberately not AI. Glass Openings is a 3D product configurator handling more than 120 products, over 70 materials and more than 30 glazing variations with live product logic; Circle Of Life is a subscription engine running order generation, billing and delivery scheduling on a workflow layer. Both are real, checkable software engineering, and the page's argument is that this is where agent builds actually fail.
What a marketing team is asked to take on trust is everything AI-specific: 200+ brands served, more than 60 per cent reduction in manual ops time "across deployed workflows", 3.4x faster lead response after rollout, and a first-agent rollout typically under two weeks. Engagements are offered as project or retainer. No individual is named.
What each agency publishes
| Agency | Named client deployment | Published price | Named individual |
|---|---|---|---|
| Agent Place | No; three agents run in its own businesses (TripleDouble, The Keyword, Wilow), and a fourth set is "in front of" Hoxton Circle | Flat monthly fee, unlimited agents; quoted on the call | Yes, Kole Ogundipe |
| Agenticise | No; three results, clients withheld | Audit from around £3,000; first build £6,000–£25,000; upkeep 10–20% of build a year | Yes, David, founder |
| flowio | Yes, The Edinburgh Boiler Company | From £750; most £1,500–£15,000; maintenance from £150/month | Yes, Malcolm, founder |
| Nectra | Yes, CK Fulfilment | Most projects start around £5,000 | No |
| Ronins | No; three builds, clients withheld | £3,000–£8,500 discovery; £12,000–£30,000 proof of concept; production from £30,000 | No |
| Tiadara | No | Typical engagements £8,000–£40,000; no fixed price stated | No |
| Turing Automate | No; a demo the page labels simulated | None | No |
| Molo Agency | Two non-AI builds described; states it has no client AI case study | None | No |
Read together, the table says something about the market rather than about any one agency in it. Five of the eight publish a price or a range. One publishes a refund term. Three name an individual. Two publish a client's own words. The test that separates them is the first one: an agency that runs agents on its own business has already solved the problem a client is paying it to solve.
Three questions the published pages leave open
What the upkeep costs, in money, in year two. Only Agenticise puts a number on it as a share of the build, and only flowio publishes a maintenance price. For everyone else it is a figure that arrives after the decision has been made.
What happens to the agent at the end. This is the one check most of the field does answer. Ronins says the code, the prompts, the workflows and the documentation belong to the client and that it trains the client's team to run them. Tiadara says the same of the code, the data and the keys, and hands over the source repositories on day one. Turing Automate says the build and its documentation are the client's and the agreement can be cancelled at any time. Molo says ownership is agreed before the work starts, either retained by the agency or handed over with the documentation and the evaluation suite. Agenticise frames its whole offer as a system the client owns rather than rents, without publishing handover terms. Agent Place publishes a term of a different kind — a monthly agreement that can be stopped in any month. Nectra publishes nothing about how an engagement stops. flowio comes closest without quite answering it: its line about a build being owned rather than rented sits under its database and app development service rather than under its automation or agent work, though it does promise full runbooks and training at handover.
Which named organisation will take the reference call. Two of the eight can point to a named client with an AI deployment written up, and a third, Molo Agency, to named clients for software work it says is deliberately not AI. Agent Place is not one of the two, on its own wording: a fourth case study names Hoxton Circle, a recruitment firm, but says the three agents built for it are "built and in front of the firm on one private page", which is a demonstration rather than a deployment. For a marketing team, that single question separates this field faster than any technical comparison.
Sources
Every claim above comes from the agency's own website, read in full on 3 November 2026: agentplace.co.uk, agenticise.co.uk, flowio.co.uk, nectra.co.uk, ronins.co.uk, tiadara.com, turingautomate.com and moloagency.co.uk. Pages were read end to end rather than sampled, because in this field the price, the maintenance terms, the exit terms and the named clients sit in a block at the very bottom. No figure here is drawn from a directory, an aggregator or an agency's own listing elsewhere. Prices and terms change without notice; the date above is the date they were read.
