The AI-first approach: how a team works when the agent does the work and people own the result

What AI-first means at Matveo

Over the past two years the phrase AI-first has turned into website decoration: it goes in the header and that is where it ends. For us it describes the order in which things get done. AI-first means that every task at Matveo goes to an agent first: setting up and running ads, mapping demand, qualifying enquiries, reports, draft copy and creatives. A person sets the goal, looks at what the agent produced and makes the call. The reverse order, where a person does the work by hand and the model offers hints, counts as a process error in our team.

This is how a team of a few people runs projects across eight markets: Russia, Cyprus, Türkiye, the UAE, Thailand, Bali and others. Without agents that volume would take a department of twenty, with media buyers, analysts and enquiry managers. With agents, each person on the team owns a whole loop: they see the demand, the ads, the enquiries and the money of one project and manage all of it through agents rather than through contractors.

It matters that AI here does not mean a chatbot on the website or a post generator. It means agents connected to the working systems, which read the data every day, make changes within set rules and bring conclusions to a person. Below we take this apart piece by piece.

Four rules the team works by

Rule one: agent first, person second. Any repeating task goes to an agent, even when doing it by hand looks quicker. A task described once then runs without a person involved, and the team's time is freed for what an agent cannot do: understand the client, negotiate, come up with an offer.

Rule two: data instead of interfaces. Agents are connected to Yandex Direct, Yandex Metrica, VK Ads and the CRM directly through MCP, an open protocol that gives a model access to tools. The agent reads campaigns and changes them through the API, instead of clicking around the ad account and waiting for a spreadsheet export. We covered how this works in detail in the article on MCP in ad management.

Rule three: every decision is measured in money. The agent sees the cost of a qualified enquiry and of a deal, not clicks and reach. It switches off ad sets that burn budget without producing enquiries and pushes the ones that lead to a call and a viewing. Optimising for clicks is banned at rule level in our setup, because a click costs nothing and means nothing.

Rule four: responsibility stays with people. Every project has the name of a team member under it. That person answers for what the agent did, the same way a manager answers for the work of their staff. The agent does not sign off the media plan and does not talk strategy with the client. It does the work that a person is prepared to answer for.

Where an agent works every day

Demand and keywords. Models tag search queries and audiences and find clusters an analyst would miss because of the sheer volume. In one luxury property project this is how the "water view" cluster was found: the buyer was searching for a view, not square metres, and the campaigns built around that query became the main source of qualified enquiries.

Advertising. The agent manages bids and budgets every day, adds negative keywords, rewrites ads with low conversion and moves budget out of burnt-out ad sets. In the morning a person gets a summary in plain words: what changed, why, and what effect to expect. The spreadsheet is attached, but decisions are made from the summary.

Enquiry qualification. An assistant replies to an enquiry within 40 seconds at any time of day, collects the budget, timeline and purpose of purchase and passes a ready, scored card into the CRM. The manager gets a conversation with someone who is buying, not with someone who pressed a button by accident. We covered how lead handling works in real estate in a separate article.

Analytics. One data model from click to contract: ads, calls, CRM and deals in a single loop. The agent builds a report for every channel all the way to the deal and explains the deviations. A person does not reconcile spreadsheets, they read the conclusions and check them.

Creative and content. Landing pages, presentations, creatives and copy are built from a brief in the client's brand style at our studio Artfkt. The designer and the copywriter work as editors: they refine, select and set the direction instead of drawing every banner from scratch.

Inside the team. The task tracker, briefs, procedures and client reports are also run by agents. The monthly report is assembled from tracker and analytics data with a single command, and a person spends their time on the commentary, not on the assembly.

What stays with people

Strategy, positioning and offers. An agent can run through a hundred options, but the choice of how the property will look in the buyer's eyes belongs to a person who has been on site and talked to the sales team.

Negotiations with the client and the sales team. Agreeing a sales plan, explaining to the owner why a channel deserves another month, resolving a conflict between marketing and sales. This is people's work, and it is where the team's value lies.

Checking the model's conclusions and making final decisions. The agent brings a conclusion and the data behind it. The person is obliged to check: look at the sample, cross-check with the CRM, ask a question. The rule is simple: the more expensive the consequences, the higher the threshold beyond which the agent must ask rather than act.

Responsibility for the result. If a campaign failed, the agent is not to blame. The person who set its rules and did not check is. We say this plainly, because otherwise AI-first turns into a way of shifting the blame onto a tool.

What the client gets from it

Speed. A change in campaigns takes hours, not a sprint until the next report. A reply to an enquiry arrives before a manager has opened the CRM. In real estate, where a buyer leaves five enquiries with different developers in one evening, that difference decides who they talk to first.

Transparency. Every action by the agent is logged: what it changed, on what data, with what effect. The client reads the same log we do and sees specific edits with reasons instead of "we optimised the campaigns". This removes the main question asked of any contractor: what did you do with my budget.

Cost. A team with no markup for size. The client pays for the system and for the people who answer for it, not for junior specialists' hours. A large agency sells hours; we sell a loop that works without growing headcount.

Where to start in your own company

Start with the one loop where lost time costs the most. Most often that is the link between the ad account and the CRM: that is where the most money is lost, because ads are optimised for enquiries, sales report in contracts, and nobody stitches the two together. One agent that reconciles spend with deal movement every morning and brings conclusions will change the quality of decisions more in a month than any new tool.

The second step is rules and thresholds. Describe what the agent may do on its own (negative keywords, bids within a range, reallocating budget inside a campaign) and where it must ask (new campaigns, budget above a threshold, switching off a channel). Without that frame the agent is either useless or dangerous.

The third step is the person who answers for it. Appoint one employee who reads the agent's summaries, checks its conclusions and makes the decisions. At Matveo we help you get there: in a five-day audit we work out which systems you have and where an agent will pay off first, and in consulting we teach your team to work with agents on its own tasks.

Frequently asked questions

What is the AI-first approach in marketing, in simple terms?It is a way of working in which any repeating task goes to an AI agent first, while a person sets the goal, checks the result and makes the decision. The agents are connected directly to ad accounts, analytics and the CRM and work every day, not on request.

Does the AI-first approach replace the marketer?No. It removes the routine from the marketer's work: reconciling reports, manual bid changes, the first reply to enquiries. Strategy, offers, negotiations and responsibility for the result stay with the person, and their role becomes closer to a manager than to an executor.

How safe is it to give an agent access to the ad account and the CRM?It is safe under the principle of least privilege: the agent gets exactly the access the task needs, works within set thresholds, and every action is logged and can be rolled back. Anything above the threshold the agent must agree with a person.

Where should a company start with AI-first?With the one loop where the most money is lost, usually the link between ads and the CRM. Then describe the rules and thresholds for the agent and appoint a person who reads its summaries and makes the decisions. Extend to the other loops once the first one is working.