I3OS: how we are making Impulsa3 agentic with AI

At Impulsa3 we did not simply decide to “use more artificial intelligence tools”. I decided that we had to redesign the way we work. I3OS, Impulsa3 Operating System, is the system I am building with the team to make our processes agentic, preserve the context of each client and turn AI into a capability shared by the whole company.

I am writing this article from my experience as CEO of Impulsa3 and as the person responsible for driving this adoption plan. This is not a theoretical guide or a promise of total automation: it is the story of why we decided to do it, which architecture we are deploying, which decisions have helped us and which limits we continue to maintain.

When I explain the difference between a tool that responds and a system that can execute work with supervision, I usually start with our analysis of AI agents in business. I3OS starts from that capability, but places it within an organisation, with processes, data and responsibilities.

The origin: an embryo that became a priority

The plan was not born in 2026 or when we started using the name I3OS. At the beginning of 2025 I launched an embryonic SEO automation pilot. We did not yet have a finished architecture or a capability inventory: we had a conviction and the need to test it through real work. The evolution of that pilot would eventually become I3OS, a system that would cover every area of the company and our organisational structure.

Iterating on this MVP, applied in practice with some of our clients, revealed that the value was not only in automating a specific task. We had built a pattern for connecting context, data, tools, method and human review. That learning opened the opportunity to do the same in other areas of the company and even in processes affecting the whole organisation.

The turning point came in December 2025. It is worth distinguishing the two dates: Anthropic officially introduced Claude Code on 24 February 2025 and announced its general availability on 22 May 2025; at Impulsa3, however, it was in December that we began to incorporate it decisively into our operations. Its ability to work with files, code, persistent instructions and tools turned the pilot into something we could evolve much faster.

Claude Code did not solve the transformation problem by itself. What it did was dramatically reduce the cost of experimenting, documenting, testing, correcting and packaging a capability. From that point on, I3OS stopped being a promising exploration and became a plan that deserved priority, investment and leadership oversight.

Why I decided we had to make Impulsa3 agentic

From my position as CEO, I could see that a marketing and digital transformation agency can grow by adding people, but there comes a point when that answer is no longer enough. The problem I saw was not only the amount of work. Each new account, service and tool increased the complexity the team had to coordinate.

I also identified a competitive risk: if we did not act, we could fall behind agencies and consultancies offering the market services similar to ours. Agentisation was not only an efficiency opportunity; it was a condition for protecting our ability to compete, maintaining quality and continuing to deliver value.

• Context becomes fragmented. A client’s information lives across folders, meetings, emails, boards, reports, analytics tools and team conversations.
• High-potential-value tasks are repeated. Researching, organising data, comparing alternatives, preparing a diagnosis or turning a meeting into actions takes time even when the method is already known.
• Quality depends too much on who executes the work. Two people can receive the same assignment and apply different criteria because the procedure lives in their experience, not in an accessible system.
• Handovers create friction. When knowledge moves from sales to projects, from projects to a discipline and from one discipline to another, context is lost and reconstruction work increases.
• Growth cannot consist of selling more hours. We needed to increase the company’s capacity without turning every improvement into a linear hire or creating dependence on a few people.

I understood that artificial intelligence could help us solve some of these problems, but adding licences without changing the way we worked would only have created more fragmentation. It would also have worsened the phenomenon of shadow AI in the company: each person using different models, prompts and criteria, without a common methodology.

My conclusion was clear: the opportunity was not to add another tool. It was to turn artificial intelligence into an organisational capability.

The message I sent to the team: adapting to grow

Agentisation did not start as an isolated technology experiment. I presented it as a company decision and wanted to explain to the team why we had to tackle it together. This was the message I sent:

“We are experiencing a historic technological disruption and, like every revolution, it brings dangers and opportunities. At Impulsa3 we have chosen which side to be on: we are going to take advantage of this paradigm shift to emerge stronger, with solidity, building together a future of sustainability and growth, both personal and professional.

That is why I challenge you to make this AI adoption plan your own. With it we will build the Impulsa3 of the future: a consultancy that drives its clients’ growth with innovative systems where technology, marketing and consulting come together, all powered by artificial intelligence.

In return I ask three things of you: focus on your daily work, an open mind and continuous training, and that you take care of and deliver your maximum value and talent to our clients and, therefore, to Impulsa3.”

Jorge Nevado Hernández, CEO of Impulsa3

With that message I wanted to make one idea clear: AI was not being implemented to replace professional judgement, but to free up capacity and raise the value we deliver. For that to happen, we needed to design a system that helped people in their daily work and evolved with them.

Why CEO sponsorship was essential

A plan of this scope could not depend on a few enthusiastic people driving it in their spare time. I decided to prioritise and sponsor it directly because, in my experience, it was the only sensible way to bring it to a successful conclusion: it needed a visible owner, an explicit priority and follow-up with the weight deserved by a transformation affecting the whole organisation.

In some previous experiences I have lived through changes of this kind first-hand, for example at Jazztel, Orange and Neoris. There I saw that organisational transformation or technology-adoption projects with a cross-functional scope often fail for a very specific reason: they are presented as important, but do not receive enough internal sponsorship, follow-up time or capacity to compete with the urgency of everyday work. The technology can work and yet the project may never become a new way of working.

But CEO sponsorship would not have been enough without Impulsa3’s internal technical profiles. Miguel Martínez, CTO, and Javier Hernández, Head of Development, have been decisive: they created and deployed the technical architecture of I3OS, connecting its different components and making it possible for the system to move from idea to operation. Without these people inside the company, we would have had to hire an external specialised service to deploy the technical side of the system, with more dependence, more cost and less capacity for internal evolution.

If the transformation affects the whole company, its sponsorship must affect the whole company too. I could not ask for complete adoption of the I3OS plan without personally taking responsibility for explaining it, prioritising it, funding it and reviewing its progress.

That does not mean the CEO has to decide every Skill or every technical detail. It means they must protect the purpose, unlock resources, demand results, listen to resistance and keep the project alive until it stops being a project and becomes the usual way of working.

Agentisation is not about filling the company with bots

When I talk about “agentisation”, I am describing the incorporation of systems capable of interpreting a goal, consulting the right context, using tools and proposing or executing a sequence of actions. But my intention has never been to hand a company over to a collection of autonomous robots.

• Automating means programming a fixed sequence so that it repeats without intervention.
• Assisting means helping a person think, write, analyse or decide.
• Making a process agentic means giving it the ability to interpret, plan and use tools within defined limits.
• Building an operating system means making that capability reusable, governable, measurable and accessible to the whole organisation.

That is why, for me, I3OS is not an application you buy and install. It is the name of the architecture and adoption plan with which we are organising Impulsa3’s AI. It can incorporate different models and tools, but it needs a common way of working.

What I3OS is and which pieces make it up

I3OS stands for Impulsa3 Operating System. It is our internal use case and the architecture with which I am driving, together with the team, the agentisation of the company. The central idea we defined is to separate what belongs to a client’s context from what belongs to a discipline’s method.

• Projects by client. Each company has a workspace with its identity, status, documents, decisions and history. This allows the system to work with context without mixing accounts.
• Cross-functional Skills. A Skill contains the method for executing a repeatable task: which sources to consult, in what order, which checks to perform, how to present the result and what a person should review.
• Data and connectors. AI must be able to reach the real information used in the work: documents, meetings, calendar, analytics, SEO, tasks and other authorised sources. A prompt is not a substitute for data.
• Governance. We define what the system can consult, which actions require approval, what information is out of scope, how a single source of truth is maintained and what cannot be delegated.
• Adoption and learning. The team proposes improvements, tests methods, communicates changes and receives training. The system grows from real work, not from an architecture designed in the abstract.

Project = the context it remembers. Skill = the method it knows how to apply. This combination allows the same way of working to adapt to each client without starting from zero.

Margin pressure forces a change in the model

The decision for I3OS adoption to reach 100% of the organisation—more than 25 people counting employees and external collaborators—does not respond to a technology trend. It responds to a business forecast: we expect margins to compress in the short and medium term in growth, marketing and technology services.

If we continue producing everything with the same model, routine and systematic tasks will consume an increasing share of the capacity of people whose real value lies elsewhere. The reduction in production cost we seek with I3OS should allow Impulsa3’s talent and experience to move towards the areas where we truly add value: understanding the client’s business, making decisions, solving complex problems, creating, innovating, supporting and taking responsibility.

We do not want talent to compete with routine. We want I3OS to absorb routine and systematic work so that people can focus where their experience makes a real difference.

That is why I do not present I3OS as an individual productivity tool. I present it as competitiveness infrastructure: if we can produce better and with less friction, we can dedicate more energy to strategy, client relationships and building solutions that cannot be replaced by generic execution.

How I3OS works in practice

1. It affects the entire client journey

I3OS has had an impact on every area of Impulsa3, from sales to production. In presales it helps us with the client’s discovery: organising available information, asking better questions, identifying needs, spotting opportunities and preparing a proposal with more context before committing to a solution.

When a new client joins the company, onboarding is no longer an isolated activity or the sum of independent pilots. We have automated and connected the pieces we tested in different pilots: reading documentation, creating context, working structure, Project, profile, logbook, data sources, initial tasks and follow-up method. This way, the learning from the pilot becomes a capability we can repeat from day one.

2. The client has its own context

The unit of work is the client company, not a standalone conversation. Its workspace brings together the context profile, living documentation, decisions, meetings and data belonging to its brands and markets. This reduces the need to explain the same thing again and prevents recommendations from being made without knowing the real constraints.

3. The task is executed with a reusable method

When we identify a repetitive task, I do not want the solution to be a secret prompt saved on one person’s computer. We turn the procedure into a Skill: a living document that defines the goal, scope, sources, steps, quality criteria and output format.

4. The system consults sources and returns reviewable work

I3OS connects reasoning to a much broader ecosystem of tools. We use Google Drive, BigQuery, Google Calendar, Trello, Gmail under an opt-in policy, Screaming Frog, Ahrefs, Semrush, Gamma, Canva and n8n, as well as a large gallery of connectors and applications hosted on private Linux VPS servers. The point is not to accumulate tools, but to put each source and capability at the service of the right process, with clear permissions, owners and limits.

I do not consider a result finished simply because the model generated it. The responsible person reviews the content, validates the data and decides what is delivered or which action is executed. This is the same logic we explain when discussing human oversight in AI: autonomy is earned through evidence, not enthusiasm.

Why I chose a Skill-based architecture

A common temptation was to build a different agent for each task or buy an already packaged “team of agents”. I chose a Skills-first architecture: first we define the capability the organisation needs and then decide whether it should live as a mode of an existing Skill, as a new Skill or as an integration.

• Avoids duplicating methodology. If two tasks share a research, validation or reporting phase, it makes no sense to maintain two incompatible versions.
• Makes knowledge portable. The procedure does not depend on the memory of the person who discovered it and can improve through team reviews.
• Allows growth in layers. First standardise the method, then connect data and finally grant more autonomy if the results justify it.
• Facilitates governance. It is easier to review a catalogue of capabilities with owners and limits than a collection of opaque conversations.

As of today we maintain an inventory of more than 80 Skills, distributed across Claude environments—Desktop, Design and Code—and ChatGPT—web and Codex. They cover production areas such as growth and ecommerce, SEO and GEO, finance, sales, technology development, operations, content, Paid Media and project management, among others. They are not more than 80 independent robots or more than 80 promises of complete automation. They are documented capabilities that help us turn repeatable tasks into a shared way of working, with different degrees of assistance and autonomy.

How we turned an SEO pilot into a company-wide system

We started with a discipline where the problem was visible and measurable: SEO. Before talking about making the whole company agentic, we tested whether we could build a client Project, load its context, connect its sources and run an audit using a common method.

The pilot forced us to solve questions that often remain hidden in a demo: how to correctly identify a company and its brands, how to preserve history, how to combine Search Console and GA4 data, how to read technical crawls, how to name documents, how to keep a logbook and how to separate a data-based recommendation from a hypothesis that still needs validation.

When the pilot worked end to end, my conclusion was not “SEO is now automated”. It was more useful: we had found a replicable pattern. We could take the same logic to client onboarding, meetings, task management, analytics, reporting and other disciplines. To evaluate impact we did not stop at time saved; we also observed consistency, reasoning traceability, data quality and the ability of a second person to continue the work.

Measurement remains essential. For each capability we want to know how long it takes, what percentage of results needs correction, how many human interventions it requires and what impact it produces for the client. It is the same discipline we recommend in our article on how to measure the ROI of artificial intelligence: do not confuse generated activity with created value.

What we have had to change in the organisation

• Documentation stopped being a dead file. Profiles, logbooks, procedures and decisions become part of the system that makes it possible to work with context.
• Meetings have continuity. Transcripts and calendars feed the client’s status and make it easier for actions not to depend on someone remembering to reconstruct them.
• Improvements have a pathway. A repetitive task or question can be proposed on an internal board; the team decides whether it deserves to become a Skill, connector or process improvement.
• Training is part of deployment. Presenting the architecture, collecting feedback and explaining progress is just as important as configuring an integration.
• The company learns while it delivers. Every real project provides cases, errors and decisions that can improve the common method.

This is also why the process cannot remain confined to the technical team. An operations person detects different friction from an SEO specialist; a project manager knows questions that do not appear in an audit; a sales profile sees product opportunities. Adoption needs all these perspectives.

From specialist departments to client squads

Implementing I3OS has also brought about an organisational change. For years we were structured around specialist areas: SEO, Paid Media, email marketing, development, design and other disciplines. This model supports technical depth, but it can create silos when a client needs a coordinated response.

We are evolving towards squads linked to our clients’ projects, with several disciplines working within the same team. Specialisation does not disappear: it serves a unit that understands the context, shares objectives and can combine SEO, Paid, email marketing, technology, content, analytics or creativity according to the problem to be solved.

This change matters because I3OS needs a clear place where context and method can be applied. The Project provides the client’s memory; the squad provides business accountability; Skills provide reusable procedures; and each specialist provides judgement. Technology connects these layers, but it does not replace the team’s conversation or decision.

The goal is not for everyone to know how to do everything. It is for the client not to have to coordinate every discipline internally to obtain a complete answer, and for us to work with a broader view, less duplication and a greater capacity to learn.

Benefits for Impulsa3 clients

The first beneficiaries of all this investment and change process are our clients. I3OS was not designed for Impulsa3 to work faster only internally; we are building it to bring each project more capacity, better quality, greater speed and a more integrated experience. Technology makes sense when it improves the results and the relationship we maintain with the companies that trust us.

1. More capacity and quality in technology

In technology, I3OS allows us to increase production capacity without lowering our standards. Skills, agents, connectors and human reviews help us document projects better, test earlier, detect errors faster and deliver more consistent solutions. This effect is not limited to the services we provide: it also feeds into our own products, such as connectors and premium Shopify apps, which we develop in our Impulsafy.io business line.

The architecture and procedures we use internally allow us to turn learning from real projects into reusable improvements for other clients and new product capabilities. The result is a shorter improvement cycle: we build, test, learn and apply again with more speed and control.

2. 360° support for ecommerce clients

For our ecommerce clients, this evolution allows us to offer 360° support in growth marketing and development on Shopify and WooCommerce. The combination of specialised Skills, MCPs and integrations within I3OS connects business analysis, SEO, advertising, CRM and email marketing, analytics, conversion and store development in the same working context.

This allows us to be much faster at identifying and implementing growth systems: identify an opportunity, prioritise it, design the solution, develop it, measure its impact and improve it without each phase being isolated in a different team. The client receives a more complete view and can move from hypothesis to execution sooner, with less friction and greater traceability.

3. Better lead-generation performance for service companies

For service companies whose growth depends on lead generation, I3OS helps us better connect value proposition, demand, acquisition and sales operations. We can analyse the potential customer’s journey more quickly, identify where intent is lost, define the architecture of pages and content, prepare campaigns, instrument forms and CRM and establish qualification and follow-up criteria.

The benefit is not simply generating more contacts. It is turning investment into opportunities with context and follow-up more effectively: knowing where each lead came from, what it needs, what response it received and which learning must return to marketing and sales. This allows us to launch experiments faster, reduce response times and make decisions with a view closer to real revenue.

4. Design, content and creativity as a cross-functional layer

In Design, Content and Creativity is where, in my opinion, the multiplier effect of I3OS is most noticeable. They are not a final phase added once everything else has been decided: they are a cross-functional layer that improves how we research, explain, present and activate every solution, from a sales proposal to an online store, campaign or content system.

• We explore more creative alternatives in less time and can prototype before investing in a complete execution.
• We adapt the message and assets to each channel, audience, stage of the journey and business objective.
• We maintain greater brand consistency when several disciplines and people are involved in the same project.
• We turn knowledge into reusable systems without losing creative direction or professional judgement.

AI brings speed and breadth; our team brings judgement, brand sensitivity and creative direction. When both capabilities work together, we can do much more for our clients and do it with a quality that does not depend on producing more purposeless assets.

Ultimately, the value of I3OS is not measured only by the hours we save inside Impulsa3. It is measured by the capacity we free up to understand the client’s business better, spot opportunities earlier, execute more precisely and support improvement from start to finish. That is why this project makes sense.

Governance: AI accelerates, but it does not sign

AI accelerates; it does not sign. Every result affecting the client, a business decision or external communication requires responsible human review.

Agentisation increases the capacity to act. That is why it also increases the need to control context, permissions and approval points. Our framework is still evolving, but there are rules that, as CEO, I do not want to negotiate:

• Purposeful access. The system should consult only the sources it needs for the task and the relevant client.
• Living sources with owners. If information is out of date or has no owner, a convincing answer may be wrong.
• Review before execution. Sensitive actions, deliverables and decisions are not delegated by default.
• Privacy by design. Personal, HR, legal or financial conversations remain outside operational workflows that do not need them.
• Explicit permissions and limits. The operational account or identity should have the minimum scope compatible with its function, and integrations should be reviewable.
• Decision records. Generating an answer is not enough: it must be possible to explain what was consulted, what was decided and who validated the result.

At this point, I3OS connects with the work we have already published on how to set up an AI committee in a company, GDPR compliance when we use AI and data quality. Experience is showing us that governance is not a document written at the end: it is a design condition.

What we have learned so far

• The bottleneck is usually context, not the model. Changing models does not fix an incomplete profile, an undated piece of data or a process nobody has defined.
• Methodology must come from real work. The best Skills are born by observing a task that is repeated and documenting how an expert performs it.
• Autonomy is graduated. We start with assistance and review; we expand actions only when we have evidence of quality and sufficient controls.
• Adoption is a people project. The team needs to understand why, practise, ask questions and see that its proposals change the system.
• Standardisation must not kill judgement. A Skill establishes the minimum common method, but the professional continues to interpret the context and take responsibility.
• A use case is more valuable when it can be repeated. Real scale appears when knowledge becomes a capability that several people can activate and improve.

What we are not promising

Telling our success story also requires explaining its limits. I3OS does not claim that a company can operate without people, that any process can be automated or that AI will be accurate simply because it has access to more data.

• We are not replacing the client relationship or strategic judgement.
• We are not giving a model unlimited autonomy to act without controls.
• We are not treating all data as if it had the same level of sensitivity.
• We are not measuring success by the number of prompts or Skills created.
• We are not hiding that the architecture has phases, pending decisions and aspects we still need to improve.

That is precisely why we prefer to show the journey: responsible implementation is more credible when it recognises limits, publishes decisions and improves through experience.

How to start making your company agentic

My experience at Impulsa3 leads me to recommend a specific path for an SME or service company:

1. Map repetitive processes. Ask which tasks consume time every week, where information is copied and where the same research is repeated.
2. Choose a low-risk, high-learning use case. Start with an internal, measurable and reviewable task, not with the company’s most sensitive decision.
3. Define the minimum context. Gather sources, owners, dates, exceptions and quality criteria before writing the prompt.
4. Document the method. Turn one person’s experience into a procedure that another person can review, execute and improve.
5. Connect only the necessary tools. Every permission and integration must have a reason, an owner and a limit.
6. Establish human approval. Define what the system can propose, what it can prepare and what it can never execute without confirmation.
7. Measure before scaling. Record time, quality, errors, corrections, interventions and value for the client or team.
8. Build an improvement loop. Every recurring failure should become an improvement to the data, Skill, connector or process.

This is the approach we are applying in our vision of moving from digital transformation to AI transformation: it is not about adding AI to old processes, but reviewing how the company should work when artificial intelligence is part of its infrastructure.

Frequently asked questions about I3OS

Is I3OS software that any company can install?

Not in the traditional sense. It is a working architecture and adoption plan based on tools, models, data and procedures. The technology can be replicated; the context, governance and Skills must be adapted to each company.

Does I3OS mean having an agent for every employee?

No. The model we are building separates each client’s context from the organisation’s common capabilities. A person can activate a Skill within the appropriate Project and receive consistent support without us having to create an isolated robot for each role.

Do we have to use a single AI tool?

Not necessarily. At Impulsa3 we use different environments depending on the work. The operational core of I3OS is being built around Projects, Skills and connectors to client context, while other tools may be better suited to an individual task, image, video or one-off exploration. The decision should follow the process and governance, not model fashion.

When will we know that a process is ready to become agentic?

When the goal, sources, steps, exceptions and quality criteria are clear enough for a second person to review them. If nobody can explain how a task is done, it is probably not ready to be delegated to a system; first we need to understand and document it.

Conclusion: our system is still learning

I3OS is the answer I am building with the team to a very specific question: how can a service company take advantage of artificial intelligence without losing judgement, context or accountability? My answer is not another chatbot, a collection of prompts or the promise of a company without people.

It is a working infrastructure: Projects with client memory, Skills with method, connected data, controls, training and a team that participates in improvement. The SEO pilot gave us the first proof. Internal adoption, the inventory of more than 80 Skills and progressive expansion into operations are giving us the learning we still need.

The most important thing for me is that this article does not describe a finished project. It describes the success story we are living at Impulsa3 and the decisions we are making to turn a technology opportunity into a sustainable operational advantage.

If you need to identify which processes in your company can become agentic, design an AI architecture with context and governance or support your team in adoption, my team and I can help you turn the opportunity into a practical, measurable working system.