“The question is no longer how big your team is. It is how well you can brief the ones that never sleep.”
I have spent more than a decade in go to market work, and I have not seen a shift this fast. The earlier changes, like marketing automation, CRMs and programmatic ads, changed how we executed. This one changes who executes. Most marketing and sales leaders I speak to now run two rosters, a human team and an agent team, and the second one is growing faster than the first.
Cold outreach has been automated end to end
In all my experience cold outreach was the most draining job in any outbound motion. I had Someone built a list, enriched it, wrote the first line, sent the sequence, read the replies, and sorted the interested from the irritated. It was repetitive and badly paid, and burnout was constant.
Agents now do the whole chain. You can set up an agent on Claude or Grok that pulls account data from sources like Apollo, LinkedIn Sales Navigator or Clay, researches each prospect from public information, writes a message that references something specific, sends through your email infrastructure, and reads the replies. It then tags each reply as interested, not now, wrong person or unsubscribe, and routes only the interested ones to a human. What used to need three or four SDRs now needs one person who designs the logic and reviews the exceptions.
The catch is that everyone has this capability. Inboxes are flooded with personalised looking messages, and buyers have learned to spot them in about two seconds. Gmail and Outlook have also tightened spam filtering and sender requirements. Automation lowered the cost of sending to near zero, which means the only thing that still works is relevance. The winners are not the teams sending the most volume. They are the teams with a sharp ICP, a real trigger for reaching out, and a clear reason the prospect should care this quarter.
Social media runs on a calendar the machine builds
Tools like Scalio, SocialBee, Buffer, Planable and Canva Magic and a growing list of similar platforms take a brand book, covering tone, colours, themes, audience and past posts, and generate a full week or month of content. That includes captions, creatives, hashtags and scheduling across LinkedIn, Instagram and X. Work that took a content manager and a designer three weeks each month now takes a few hours of review.
This is useful for consistency, which is where most small brands fail. Most companies do not lose on social because of bad content. They lose because they disappear for six weeks. An automated calendar fixes that.
The risk is sameness. When a thousand brands in the same category feed similar prompts into similar models, the feed fills with similar posts. The brand book becomes the real asset. If yours is vague, your automated content will be vague at scale. Spend the time on the inputs, and keep a human responsible for the two or three posts a month that carry an actual opinion or a real customer result.
Paid media has become a conversation with the algorithm
Meta Advantage+ is the clearest example. You supply creatives, a budget, a goal and some audience signals, and the system decides placement, audience, creative combinations and bidding. Manual audience targeting, which used to be a core skill for media buyers, now matters much less. Google has moved the same way with Performance Max and its AI driven campaign types.
Two things follow from this. First, the skill of a media buyer has shifted from setting up targeting to feeding the system good creative and clean conversion data. If your pixel or conversion API is sending poor signals, the algorithm optimises toward the wrong customers, quickly and at scale. Second, you lose some transparency. These campaigns often give you less detail on exactly where the money went, so you need your own measurement outside the platform, such as holdout tests and blended cost per acquisition against actual revenue.
Search is splitting, and ads are following the audience
A growing share of product research now starts inside AI assistants rather than a search box. People ask ChatGPT, Gemini, Perplexity or Claude for the three best tools for a task and treat the answer as a shortlist. If you are not in that answer, you were not considered.
Ad platforms are chasing this behaviour. OpenAI has been building advertising into ChatGPT, and early advertisers are watching it closely because the intent in a conversation is often more specific than a keyword. Google already shows ads inside AI Overviews and AI Mode in search, and how far advertising extends into the Gemini app itself is something every marketer should track closely. I would be surprised if a full ad product inside conversational AI is not standard across the major platforms within the next year or two. Verify current availability and formats before you plan budget around any of them, because this space changes month to month.
What you can do today does not depend on ad products. Treat your visibility in AI answers as a channel. That means clear product pages, comparison content, structured information, third party reviews on sites like G2 and Capterra, and consistent mentions across credible sources. Some call this generative engine optimisation or answer engine optimisation. The name matters less than the habit of checking, every month, what these assistants say about you and your competitors for your top twenty buying questions.
Market research is mostly desk research now, but authenticity is the open question
I remember the days when market research meant going door to door. On a new product launch I would be out there myself, and my team would be too, asking people questions face to face. We reached our conclusions from the survey sheets and a good deal of market gut feel. The samples were small, a few hundred responses on a good round, but there was authenticity in it. You saw the reaction on a person’s face, you heard the pause before an answer, and you knew the person in front of you was real.
Today a capable marketer can build a market landscape in an afternoon. The deep research modes in Claude, ChatGPT and Gemini read thousands of sources, and the same goes for audience sizing, reach and frequency estimates for above the line and below the line planning, and channel benchmarks. That is a scale no field team could match, and it takes out the personal bias of whoever happened to run the survey. It does not take out bias altogether, though. It moves it. The model carries the bias of its training data and of the sources it could reach, and those lean towards businesses that are well documented online. The small town buyer you are trying to understand is rarely one of them.
Primary research has a bigger problem. Online surveys are now being filled in by bots and AI agents chasing the incentive, and panel providers have started to say so openly. Synthetic respondents, where a model simulates how a target segment would answer, are being sold as a substitute for real research. They can sharpen a hypothesis, but they cannot tell you what a real buyer will do with their own money. My rule is simple. Use AI for desk research and to form hypotheses, then test them in conversations with real customers and with prospects who chose a competitor. Ten honest interviews will tell you more than a thousand doubtful survey responses.
The two team model and the cost of managing it
Most teams I work with now have a human roster and an agent roster. The agent roster handles the work nobody enjoyed. It does daily task follow ups, reminders to account owners, CRM hygiene, meeting notes, weekly pipeline summaries and first draft reports. Grok, Claude and the agent features inside tools like HubSpot and Salesforce are all being used this way.
Managing an agent team is a real skill. Agents need a clear brief, defined limits on what they can send or change, a log you can audit, and an owner who is accountable for the output. The mistakes are expensive when they happen, like a wrong message sent to a hundred customers or a pipeline report built on bad data. Treat each agent like a new junior hire with unlimited stamina and no judgement. You would not give a new hire access to send anything to your whole customer base on day one.
What this means for GTM leaders
Here is what I would do if I were running a GTM function this quarter.
- Map every recurring task in your sales and marketing process and mark which ones are rules based. Those are candidates for agents. Anything that needs judgement or relationship stays human.
- Fix your inputs before you automate. That means your ICP, brand book, messaging, CRM data and conversion tracking. Automation multiplies whatever quality you start with.
- Build measurement that does not depend on any one platform, so you can tell whether an automated channel is really producing revenue.
- Audit your presence in AI answers monthly and treat it as a channel with an owner.
- Keep a human on every point where a buyer forms trust, which means the first real conversation, the proposal, the negotiation and the customer reference.
- Hire for judgement and taste. The people who will matter are the ones who can brief, review and correct the agents, and who understand the customer well enough to know when the output is wrong.
The part that has not changed
AI has removed a large amount of manual effort from GTM, and the teams that adopt it well will run leaner and faster. But the fundamentals are untouched. You still need a clear segment, a real problem worth solving, a message that fits it, and a product that delivers. Automation makes a good strategy scale quicker. It makes a weak one fail quicker and at a higher volume.
The companies that come out ahead will not be the ones with the most agents. They will be the ones that know exactly what they are asking those agents to do.
