Let me be uncomfortably honest for a moment. The thing I do for a living, running a marketing agency, is being reshaped by AI faster than almost anyone in the industry is willing to admit publicly.
Not the vague, hand-wavy "AI will change everything" that people have been saying since ChatGPT launched. Specific, practical, already-happening change. AI agents, autonomous software systems that can execute multi-step marketing tasks without constant human input, went from research papers to production tools in about eighteen months. And they're getting better every quarter.
I use them in my own business. Every day. And I think anyone running a B2B marketing operation who isn't paying attention to this is going to be in trouble sooner than they expect.
An AI agent is different from a chatbot. A chatbot answers a question. An agent completes a task. That distinction matters enormously.
Right now, in September 2025, AI agents can reliably do the following things.
Run email sequences. Not just schedule pre-written emails, but draft contextually appropriate follow-ups based on recipient behaviour. If a prospect opened email two but didn't reply, the agent drafts a follow-up that references the content of email two and suggests a specific next step. The drafts still need human review, but they're 80% there.
Score and prioritise leads. Agents can analyse a CRM full of contacts and rank them by probability of conversion based on engagement history, firmographic data, and behavioural signals. A task that used to take a marketing manager half a day now takes an agent about forty seconds.
Build campaign reports. Pulling data from multiple sources (email platform, CRM, analytics, ad platforms), combining it into a coherent narrative, highlighting anomalies, and suggesting actions. The reports aren't perfect, they miss nuance that an experienced analyst would catch, but they're a solid first draft.
Research prospects. Given a company name, an agent can find recent news, identify key personnel, summarise their LinkedIn activity, and flag potential pain points relevant to the product you're selling. This used to take a junior researcher 20 minutes per company. The agent does it in 90 seconds.
Draft responses to inbound enquiries. When a lead fills in a contact form, an agent can draft a personalised response within minutes, referencing the specific service the lead asked about and suggesting relevant times for a call. These drafts have roughly a 70% acceptance rate from the human reviewer. The other 30% need meaningful editing.
The trajectory is steep. Based on what I'm seeing in development and what's already in beta, here's what I expect by early 2026.
Agents will manage entire outreach sequences from data identification through email sending to meeting booking, with a human only approving key decision points rather than reviewing every output. The human becomes a quality gate, not a doer.
They'll generate and test creative variants at a speed that makes traditional A/B testing look quaint. Instead of testing two subject lines, you test twenty, each tailored to a different segment, with the agent automatically shifting volume to the best performers.
They'll monitor campaign performance in real time and make adjustments without waiting for a weekly review meeting. If an email campaign starts underperforming on Tuesday, the agent adjusts messaging, pauses underperforming segments, and reallocates budget to what's working, all before Wednesday morning.
They'll handle multi-channel orchestration: coordinating email, LinkedIn, phone outreach, and content distribution as a single, synchronised system rather than separate channels managed by separate people.
This is the part that matters most, and it's where I think a lot of the panic is misplaced.
AI agents won't replace relationships. When a client calls me because they're worried about their pipeline, they don't want to talk to an agent. They want to talk to someone who understands their business, their market, their specific anxieties, and can give them honest advice. That conversation requires empathy, context, and years of experience. Agents have none of those.
Agents won't replace judgement. They can tell you what the data says. They can't tell you whether the data is asking the right question. An agent might recommend doubling down on a campaign because the open rates are high, while a human recognises that the opens are coming from the wrong segment and the campaign needs redirecting, not amplifying.
They won't replace industry knowledge. Understanding that a seal manufacturer's buying cycle peaks in Q1 because that's when maintenance budgets reset, or that accountancy practice sales slow during tax season, requires contextual knowledge that comes from working in a market, not from processing data about it.
And they won't replace the ability to say "actually, we shouldn't do that." Agents optimise toward the goal you set. If the goal is wrong, the optimisation is wrong. Someone needs to set the strategy, challenge the assumptions, and occasionally override the system. That's a human job.
Here's where my thinking might differ from the conventional wisdom.
Large agencies are threatened by AI agents because their business model depends on headcount. A 50-person agency billing for 50 people's time can't easily justify those fees when an agent does the work of ten of those people. The overhead, the office, the middle management, the account executives, all of that becomes harder to justify when the production work is increasingly automated.
Small agencies, particularly one-person or small-team operations, have the opposite problem. They've always been constrained by capacity, not overhead. AI agents remove that constraint. A solo operator with the right tools and the right agent infrastructure can now deliver what a 15 to 20-person agency delivered three years ago. Same quality data builds. Same campaign execution. Same reporting depth. Fewer people, lower cost, faster delivery.
The economics shift in favour of lean operations with deep expertise and strong technology. The value isn't in having a building full of juniors doing manual work. The value is in knowing what to build, why, and for whom. Strategy, relationships, and judgement become the differentiators. Everything else is execution, and execution is increasingly automated.
I'll tell you exactly how I use AI in my business. Data research and enrichment: heavily assisted by AI. Email draft generation: AI produces first drafts that are always human-reviewed and usually rewritten significantly. Campaign reporting: AI pulls the data and structures the report, I add the analysis and recommendations. Content outlining: AI helps me organise thoughts and identify gaps. I write the actual content.
What's never AI: strategy conversations with clients. Final sign-off on anything that goes out. The phone calls Rebecca makes to directors. The relationship management that keeps clients long-term. The judgement about what a specific business needs versus what the data suggests.
I don't hide this from clients. If anything, they appreciate the honesty. They'd rather know that AI helped build their prospect list of 11,400 contacts in three weeks instead of three months, and that a human verified every email and scored every contact, than pretend it was all done by hand.
The agencies that pretend AI doesn't exist in their workflow are either lying or falling behind. The agencies that use it as a crutch and skip the human quality layer are producing mediocre work. The sweet spot is using agents for what they're good at, speed and scale, while keeping humans where they matter, strategy and quality.
That's the model. It works. But it requires honesty about what the technology does and what it doesn't. I'd rather have that conversation openly than wait for a client to figure it out themselves.
Martin Dugan, AA2
