AI agents for marketing, and what they can do today
Short answer
An AI marketing agent is software built on a language model that works towards a goal by choosing its own steps and using tools, such as searching the web, querying a contact database or writing to a CRM. Agents already do narrow research and data jobs reliably. They struggle with open-ended goals, so the useful ones have a tight job, firm limits and a person checking their work.
By Jakub Cambor, founder of AI for Marketing, updated
What is an AI marketing agent?
An AI marketing agent is a program built on a language model that is given a goal and a set of tools, and decides for itself which steps to take: which searches to run, which records to read, what to write back and when it has finished. Anthropic, the company behind Claude, draws the line this way: workflows orchestrate models and tools through predefined code paths, while agents direct their own process and tool use.[1]
| Compared on | Chatbot | Workflow with AI steps | AI agent |
|---|---|---|---|
| What starts it | A person typing a question | A trigger such as a new form or a schedule | A goal, handed over by a person or a trigger |
| Who decides the steps | The person asking | Whoever built it, in advance | The model, as it goes |
| Tools it uses | Usually none, or web search | The ones wired into each step | Any it has been given, chosen as needed |
| Marketing example | Answering product questions on a website | A weekly report pulled together and summarised | Researching target companies and finding the right people in them |
| Main risk | Wrong answers given to customers | A step breaking on input nobody expected | Errors that compound across steps, and higher running costs |
In marketing the word gets stretched. Many products sold as agents are chatbots or fixed workflows with a new name, which is not a problem in itself, because a fixed workflow is often the better tool. Anthropic's own advice is to find the simplest solution that works and add complexity only when it is needed, since agentic systems often trade speed and cost for better task performance. Our article on AI agents and chatbots goes further into where each fits.[1]
Interest is far larger in the US than in the UK, and it centres on agents for business in general rather than on agencies that build them.
| Search | UK | US |
|---|---|---|
| ai agents for business | 110 | 3,600 |
| ai marketing agent | 390 | 1,900 |
| ai agents for marketing | 70 | 480 |
| ai agent agency | 10 | 70 |
| All of these | 580 | 6,050 |
Monthly Google searches for each phrase, in each country.
Source: Google Ads monthly search volume, via DataForSEO (September 2026)
What can AI agents do for a business today?
Today AI agents reliably do narrow jobs with a clear finish line and output a person can check: researching a list of companies, finding and verifying the right contacts in each, sorting an inbox or a support queue, and drafting content from a brand's own material for someone to approve.
Allegiance Industries, a $100M+ facility services company, is a working example. Its sales reps used to research prospects one at a time across LinkedIn, company websites and directories, an estimated 15 to 25 hours a week. The system built for it in 2026 runs every morning, and an agent in the middle makes the research decisions a rep used to make.[3]
- 1
Google Sheets
The sales team adds target companies to one sheet for manufacturing and one for education.
- 2
Make.com
A scheduled scenario reads the new companies each morning, passes them to the agent in turn, retries failures and logs every run.
- 3
Relevance AI agent
It decides which job titles to look for in that sector, builds the searches, checks the contacts that come back and assembles the record.
- 4
Apollo
It supplies the contact data: email, direct phone, LinkedIn profile and company details.
- 5
Zoho CRM
The finished lead lands with 17 fields filled, after a duplicate check; a company with no valid contacts is flagged for a person to review.
Source: Automated lead prospecting: AI for Marketing case study
Up to 41
Automated lead prospecting
Up to 41 companies researched per daily run, 17 data points per lead, and an estimated $9,000 to $10,000 a year saved (the client's own estimate).
Allegiance Industries, a $100M+ facilities company, built by AfM in 2026 with Make.com, Relevance AI, Apollo and Zoho CRM.
Read the case studyNotice what makes it work. The agent has one job, a fixed set of tools, a clear test of a good result, and a rule for what to do when it is unsure. That shape is what separates the agents that stay in use from the ones that get switched off. Where agents still fail is just as predictable:
- Open-ended goals: an agent told to grow your Instagram has no finish line and no test of success.
- Long chains of steps: Anthropic warns that the autonomy of agents brings higher costs and the potential for compounding errors.
- Untrusted input: an agent that reads emails, web pages or reviews can be steered by text planted in them, and prompt injection tops the OWASP list of risks for language model applications.
- Spending or publishing: anything that moves money or speaks for the brand in public needs a person to approve it.
Should I hire an AI agent agency?
Hire an AI agent agency when you can name the job the agent will do, the number it should move and the person who will check its work. If you cannot yet name those three, start by finding the job, because an agent built for a vague goal is the kind of project that gets cancelled.
The cancellations are expected to be common. Gartner predicted in June 2025 that over 40% of agentic AI projects will be cancelled by the end of 2027, because of escalating costs, unclear business value or inadequate risk controls. It also warned about agent washing, the rebranding of assistants, RPA and chatbots as agents, and estimated that only about 130 of the thousands of vendors claiming agentic AI are real.[5]
- Ask to see an agent it built running, with the logs, not a demo video.
- Ask when it would build a fixed workflow instead; a good builder turns some agent requests down.
- Ask what the agent does when it is unsure. The answer should be a review queue, not a guess.
- Ask who owns the accounts, prompts and data, and what you keep if you stop.
- Ask for the monthly running cost, split into tools and model usage.
We build agents where the job calls for one, as part of running a client's marketing or as an add-on to it, and we would rather build a plain workflow when that does the job. Jakub has been a certified Relevance AI partner since April 2025. You can read the Allegiance Industries case study for the full build, including the client's own account of it.[6]
Proof from our own work
1,600+
Inbound lead automation
1,600+ leads processed in the peak quarter; follow-up went from 1 to 3+ days to the same day.
Global technology platform (anonymised), in the same engagement as the paid media result: lead capture, enrichment and CRM entry automated. Delivered by Jakub as a freelance consultant before or while founding AfM.
Read the case study4
Brand-trained SEO content agent
An agent with 4 capabilities built for a premium sake brand, delivered to the client.
HEAVENSAKE, a premium sake brand. Delivered to the client, not yet deployed on their site.
Read the case study
What we do
We run your marketing for you and make it pay for itself.
- First comes a free written breakdown: we look at your site and any numbers you choose to share, with no logins, and show you where the money leaks and the first thing we would fix.
- Then we fix and run the connected commercial journey: the work that earns attention, the pages that turn it into a lead or sale, and the follow-up that creates the next valuable conversion.
- Everything is actively managed and improved, with a weekly report showing what changed and what it made, and you approve everything before it goes out.
- Rolling monthly. No lock-in, cancel any time.
What it is not
- Not a single-channel agency: we do whatever makes you money next, not ads forever or SEO forever.
- Not a junior team on a six-month lock-in: you work with Jakub directly, rolling monthly.
- Not AI left to run on its own: every piece is reviewed by a person and approved by you.
We also build custom automations, such as reporting digests, lead enrichment and routing, and support triage in tools like n8n, Make and Zapier. They are an add-on to the managed service, not a separate offer.
More questions
What does it cost to run an AI agent?
Each run costs a little in model usage, and the tools around the agent carry their own subscriptions: the Allegiance system uses Make.com, Relevance AI, Apollo and Zoho CRM, each on its own plan. Agents usually make more model calls than a fixed workflow doing the same job, so ask for the running cost per month at your volume before you commit.[3]
Can an AI agent run my marketing on its own?
No. Agents are good at bounded research and data jobs inside a process, and poor at owning a goal like growth, where the judgement about budget, offer and brand sits. Every marketing agent worth having still hands its work to a person before anything reaches a customer.
Related answers
- AI automation agencyAn AI automation agency designs, builds and looks after workflows that connect your business tools and use AI models for the steps that need reading or judgement, such as summarising, classifying or enriching.
- AI lead generationB2B companies use AI for lead generation in four places: researching which companies and people fit, enriching each lead with verified contact and company data, drafting outreach that refers to something real about the prospect, and routing new enquiries to the right person at once.
- AI consultant and AI implementationAn AI consultant finds the jobs in a business where AI can save time or earn money, builds the tools or automations that do them, and measures whether they worked.
- AI marketing automationAI marketing automation is marketing automation with an AI model inside the workflow, so a step can read, judge or write instead of only following a fixed rule.
Sources
- 1.Anthropic, Building effective agentsanthropic.com
- 2.Google Ads monthly search volume, via DataForSEO (September 2026)dataforseo.com
- 3.Automated lead prospecting: AI for Marketing case study
- 4.OWASP, Top 10 for Large Language Model Applications (2025)genai.owasp.org
- 5.Gartner, Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027 (June 2025)gartner.com
- 6.About AI for Marketing and Jakub Cambor