How to implement AI agents in 90 days

Quick answer

Implementing AI agents in 90 days requires splitting the project into two phases. In days 1 to 30, the company builds the foundation: senior leadership buy-in, selection of pioneer areas, definition of strategic objectives and project governance, appointment of an AI Officer and AI Leaders in each pioneer area, mapping of pain points, and an agent development plan.

In days 31 to 90, execution begins: running the roadmap, putting the first agents into production, creating internal learning routines, reporting results to the C-Level, and promoting internal cases.

The roadmap comes from Renato Ferreira, CRO of Tess, in an interview with the TheAgent portal. According to him, the error that stalls most projects comes before technology: confusing using AI with implementing AI.

What does implementing AI agents mean (and not just using AI)?

Implementing AI agents means changing how work gets done, with an owner, a process, and measurement. Using AI is individual: a person opens a window, writes a prompt, and improves their own output.

That distinction is the basis of the entire 90-day roadmap. In Renato's words:

"Using is individual and does not scale. Implementing is changing how the work is done, with an owner, a process, and measurement, and it is a collective, collaborative effort."

An AI agent, in this context, is a digital worker with a defined function: it has an objective, working instructions, persistent memory, access to tools through connectors, and its own execution, without depending on someone to ask.

What is the 90-day roadmap for getting AI agents up and running?

The roadmap has two phases of unequal length and distinct purpose: 30 days to build the institutional foundation of the project and 60 days to execute and prove value.

Source: Renato Ferreira, CRO of Tess, in an interview with TheAgent.

Days 1–30: what to do in the foundation phase?

The foundation phase exists to turn AI into an institutional project, with a name, an owner, and a mandate. Without that, nothing else happens. There are five moves.

1. Senior leadership buy-in. Leadership defines the strategic objectives, grants the mandate, establishes governance, and protects people's time to experiment.

2. Definition of pioneer areas. Instead of announcing a generic corporate transformation, pick a few areas to start with.

3. Strategic objectives and project governance. This is where permissions, which data can and cannot circulate, who approves what, and how results will be measured are settled. Governance defined on day 10 prevents a project stalled on day 60 by legal or information security.

4. Appointment of an AI Officer and AI Leaders. The AI Officer answers for the project as a whole. The AI Leaders are references inside each pioneer area, people who know the process from the inside. Renato is categorical about the profile of this leadership: it must always be a business leader, never IT.

5. Mapping pain points and the agent development plan. The mapping becomes a prioritized list of agents to build, with an owner and a sequence.

One detail frequently forgotten in this phase: measuring the baseline. Without the number from before (response time, volume handled, cost per task), there is no way to prove the after.

Days 31–90: how to execute and put the first agents into production?

The execution phase has a simple goal: agents genuinely running, every day, with results visible to the C-Level. Four fronts advance in parallel.

1. Execute the roadmap and put the first agents into production. Production means the agent executing recurring work, not a pilot running in a test environment.

2. Build internal learning routines. Short, recurring sessions where each area shows what it built, what failed, and what it reused. This is the mechanism that turns individual knowledge into collective capability.

3. Report results to the C-Level. Recurring reporting to leadership is what keeps the project alive in the budget and on the company's agenda.

4. Internal case promotion program. According to Renato, internal cases convert better than any external training. Seeing the colleague at the next desk solve a real problem with an agent moves more people than a workshop.

What is the most common mistake in enterprise AI adoption?

The most common mistake is confusing using AI with implementing AI. The company buys licenses, watches login charts climb, and concludes it is adopting AI. In Renato's reading, at best it is paying for people to use AI better on their own.

He describes the pattern he sees among founders:

"I keep hearing that famous line from founders: 'I'm a heavy user of Claude.' Claude is the best tool in the world for individual use, but it is not good for implementation in companies."

The point is not model quality. It is design: tools built for individual use were not made for collective use. What happens in practice, Renato says, is that the founder and half a dozen heavy users create excellent things, while the company as a whole still has no implementation process.

What are the symptoms of an AI project that will never ship?

Renato lists four symptoms that repeat in stalled companies. They work as a diagnostic checklist before starting the 90 days.

• A PoC that never reaches production. The pilot works, gets presented, applauded, and shelved.
• An initiative centralized in IT or Innovation, with no involvement from the business areas. The person who truly knows the process is the one running it every day, and that person is not in the room.
• No baseline measured before starting. Without a starting number, any result is an opinion.
• Nobody with a first and last name answering for the result. Diffuse accountability is the same as no accountability.

How do you keep the project from becoming a committee and a slide deck?

The way out, according to Renato, is turning AI into an institutional project: governance, methodology, and a tool designed to be used by everyone, collaboratively. And leadership of the project should sit with a business leader, never with IT.

On the temptation to concentrate the topic in the executive suite, the title of his interview sums up his position: no AI committee has ever automated a process. When AI stays only with leadership, the output is a committee, a roadmap, and a presentation.

This does not mean the absence of leadership. It means a clear division of roles: leadership defines objectives, mandate, and governance; execution happens in the areas, with the people who know the process.

When the 90-day roadmap is not the right path

Strategic honesty matters more than urgency. The roadmap assumes three conditions. If they are not there, 90 days tend to produce frustration.

• There is no real mandate from senior leadership. Without buy-in from the top, the AI Officer becomes a role without decision-making power. In that scenario, the prior step is convincing the board, not opening the roadmap.
• There is no process with a queue. If the company has no mapped repetitive volume, start by measuring processes before automating them.
• The priority is the most strategic, most complex case. Start with the most repetitive one instead: credibility is built with delivery, not only with a roadmap.

One caution about expected outcomes as well: 90 days delivers the first agents in production and evidence of value. It does not deliver an entire company running on agents.

Why run this roadmap on Tess?

The 90-day roadmap fails for two operational reasons: the tool was not built for collective use, or the pricing model penalizes expanding access. Tess was designed to solve both.

The platform brings together more than 250 AI models for text, image, audio, video, music, and avatar in a single environment, with shared memory and connectors to the systems the company already uses, via API or Zapier. Each pioneer area builds its own agents across multiple steps, without depending on an IT queue, which is what sustains execution by the people who know the process.

On the governance side, leadership defines permissions, restricts access, and sets budgets per person, per agent, and per team. That is what makes it possible to give areas autonomy without losing control, exactly the balance the foundation phase needs to design between days 1 and 30.

And for hybrid work between people and agents, Cowork gives the company visibility into what each employee and each agent is executing. Anyone on any plan also gets access to AI University, the beginner-to-advanced training program that supports the internal learning routines of the execution phase.

It is the difference between having an elite tool and having an elite operation.

Frequently asked questions

Is it possible to put an AI agent into production in 90 days?

Yes. In the roadmap from Renato Ferreira, CRO of Tess, the first agents reach production during the execution phase, between days 31 and 90. The condition is that the pioneer area builds its own agent for a pain point already mapped in the foundation phase.

Who should lead an AI agent implementation project?

A business leader, never the IT department, according to Renato Ferreira. The recommended structure is an AI Officer answering for the project and AI Leaders in each pioneer area, people who know the process from the inside. IT participates as a partner for integration and security, not as the owner of the result.

What is the difference between using AI and implementing AI?

Using AI is individual and does not scale: each person improves their own output. Implementing AI means changing how the work is done, with an owner, a process, and measurement, as a collective and collaborative effort. A rising login chart measures usage, not implementation.

Why doesn't an AI committee solve the problem?

Because a committee produces a roadmap and a presentation, not an automated process. Execution has to happen in the business areas, with the people who know the process from the inside. Leadership's job is to define objectives, mandate, and governance.

What is an AI Officer?

It is the person appointed to answer for the company's AI agent implementation project, with a first and last name. In the 90-day roadmap, that appointment happens in the foundation phase, alongside the appointment of AI Leaders for each pioneer area.

Start with the foundation

The 90 days do not start with choosing a model. They start with a mandate, an owner, and a process that already has a queue.

If you are designing that plan now, explore the Tess platform or talk to the sales team to map the pioneer areas in your company.