Free auditGet your free creative audit

How we work

A day in the life of an AI-powered agency

We rebuilt our agency around AI agents and one shared data connection. Here's what an ordinary Tuesday actually looks like now — told by the engineers who built the machinery and the strategists who stopped needing to think about it.

Written by
The Consigliere team
Published
Reading time
6 min

The report is already waiting

When Amy, one of our creative strategists, opens her laptop at half past eight on Monday morning, the week's performance report for her account — a Fortune-500 home retailer — is already sitting in the channel. Not a spreadsheet export someone stayed late to assemble: a written analysis of every ad that ran last week, what beat its targets, what didn't, and a short list of things worth a human's attention. It was generated overnight by one of our reporting agents, checked line-by-line by a second agent against a quality checklist, and only then delivered. Nobody at Consigliere builds Monday reports anymore. We read them. That single change is the easiest way to understand what "AI-powered agency" means in practice: the hours that used to go into assembling information now go into acting on it.

Amy · Creative Strategist

"My old Monday started with four browser tabs and an excel table. Now it starts with a paragraph that says 'these two ads are fatiguing, this hook is outperforming, here's the anomaly worth looking at.' I spend the morning deciding, not collecting."

Monday morning, before and after

Illustrative · typical account

Time from opening the laptop to the first performance decision, per account

  • Assembling data
  • Human judgment
Rounded, representative figures for a typical mid-size account. The gray disappears not because the work vanished, but because agents do it before the workday starts.

The plumbing nobody sees: one connection to everything

None of this works without a piece of internal tooling most of our clients never hear about: our MCP — think of it as one universal socket that plugs our AI into everything we know. Every ad result, every competitor creative we've cataloged, every customer review, every meeting transcript flows into one warehouse, and the MCP is the standard connector that lets any agent — or any teammate, in plain English — ask questions of all of it at once. Emiliano, who manages client relationships and wouldn't describe himself as technical, uses it the way most people use a search bar.

Alex · Engineer

"Before the MCP, every answer lived in a different tool, and only a few people in the company knew where to find it. Now the requirement is simple: if data exists, it lands in the warehouse, and the MCP exposes it. A few of us used to be the bottleneck all day — now everyone is empowered to find the answer quickly, themselves."

Emiliano · Creative Strategist

"I typed 'which of our beauty client's ads mention scent, and how did they perform?' — a question I'd have been embarrassed to ask an engineer for. I had a table in the example media files in thirty seconds. That was the moment this stopped feeling like engineering's project and started feeling like mine."

One socket, every question

Simplified

What our MCP actually does: everything we know, answerable from one place

MCP stands for Model Context Protocol — an open standard for connecting AI to data. The detail that matters: agents and people query the same source of truth, so no one works from a stale copy.

Research that never sleeps

Late morning, Emiliano opens the weekly organic research digest. Every week, our research agents sweep public ad libraries and organic social content across our clients' categories — cataloging thousands of competitor creatives, scoring which hooks and formats are gaining traction, and pulling the raw customer language from reviews that ends up in our ad copy. A strategist browsing by hand has taste but no coverage; an agent has coverage but no taste. The digest is where they meet: the agent reads everything so that Amy and Emiliano only look at the twenty examples per week that deserve human eyes.

Alexa · Client Partnerships

"Clients used to ask 'what's our competitor doing?' and the honest answer was 'whatever I happened to scroll past.' Now I answer with a number and three examples. It changed the temperature of those conversations completely."

What one week of competitive research covers

Illustrative · per client category

Competitor creatives reviewed per week — and what a human actually needs to look at

Rounded, representative volumes. The point isn't the big orange bar — it's the small green one: more coverage and less to read, because the agent filters as well as gathers.

Closing the loop: every ad remembers why it exists

Early afternoon is when the newest part of our system earns its keep — what we call closing the loop. Every ad we ship now carries its own birth certificate: the hypothesis it was built to test ("customers in this segment respond to durability claims over style claims"), the research that inspired it, and — once results come in — the verdict. Strategists have always had these hypotheses in their heads; nobody has time to write them down at volume. So we flipped it: an AI service looks at each finished ad, drafts the hypothesis and research summary itself, and Amy corrects it rather than writing it. The corrections train the system; the verdicts accumulate into an institutional memory that gets smarter with every ad any client runs.

Simon · Engineer

"The requirement from the business side was blunt: prove which creative ideas actually work, per brand, with receipts. You can't do that retroactively. So the engineering answer was to make the paper trail automatic — AI drafts, human corrects, result attaches itself."

Jordan · Creative Director

"I was skeptical — it sounded like homework. But correcting a draft takes me thirty seconds, and three months later I can ask 'what have we learned about value messaging for this brand?' and get an answer backed by every ad we've ever run. That's the thing I always wished agencies had."

The loop, closed

Simplified

Each ad's idea, evidence, and outcome stay connected — so learning compounds instead of evaporating

"Which ideas actually work for this brand?" — finally answerable with receipts

Before this loop existed, the hypothesis lived in a strategist's head and the result lived in a dashboard — and they never met. Now they're two fields on the same record.

What the engineers actually do all day

A reasonable question: if agents write the reports and sweep the research, what does our engineering team do? The honest answer is that they've become managers — of a workforce that happens to be software. Every agent we run is treated like a promising junior employee: it gets a written job description, its work is reviewed before it reaches anyone important, and it earns autonomy gradually. Our standard is the report-check pattern: one agent does the work, a second agent audits it against an explicit checklist, and a human spot-checks the audits. Requirements flow the other way, too — when the business side says "clients need this weekly, in this format, ready when they sit down Monday at 8:30," that sentence becomes an engineering spec by Tuesday.

Martin · Engineer

"People imagine we spend our days prompting a chatbot. We spend them the way any good manager does: writing clear briefs, building review systems, and deciding which work is ready to be delegated. The craft moved up a level — from doing the task to designing how the task gets done reliably."

Simon · Engineer

"Nothing ships to a client on autopilot. Campaign uploads, for instance, land in a paused state — the agent does the tedious part perfectly, and a human presses launch. That's a design principle, not a limitation."

The rule we run the company on: agents do the work that scales; humans make the calls that matter. Every system we build is an opinion about where that line sits — and the line moves only when an agent has earned it.

How an agent earns trust

Our standard pattern

The report-check pattern: nothing reaches a client with fewer than two reviews

As an agent's track record grows, human review narrows from every item to a spot-check — it never disappears.

The same pattern now runs our reporting, research digests, and campaign uploads. New agents start with full human review and earn their way to spot-checks.

What this means if you work with us

Here's the part that matters if you're a brand deciding who to trust with your growth. Because the machinery already exists, a new client isn't a cold start: once we have access to your historical data, our target is to have it flowing through the warehouse, queryable, and feeding your first reports within 36 hours — not the weeks of "onboarding" the industry has taught you to expect. From day one you inherit the same defaults our longest-standing clients have: weekly reports that write and check themselves, research agents sweeping your competitive landscape, and a memory that records why every ad existed and what it taught us. And because agents absorb the repetitive work, the humans you're paying for — the strategists, the creatives, the engineers — spend their hours on the one thing software still can't do: judgment about your brand.

The day, replayed

A typical Tuesday

What happened by itself (agents) and where the humans came in (judgment)

Orange happens whether or not anyone is at a desk. Green is why you hire an agency at all.

The honest summary of the whole system: the day starts with answers instead of questions, and ends with next week already in motion.

Your move.

We built this system for ourselves first, and run it every day on our own accounts before any client's. If you'd like to see what your Monday-morning report would look like — with your data, your competitors, your customers' words — we'll show you the real thing, not a slide about it.

Typical time from first data access to first live report: 36 hours.

Book a demo

Team quotes are drafted from internal interviews and planning notes. Charts marked "illustrative" use rounded, representative figures rather than audited metrics. Client engagements are described without names by design. MCP (Model Context Protocol) is an open standard for connecting AI systems to data sources.

All articles