Agentic AI and AI Agents for SMEs: when hiring more people is no longer the answer

Gonçalo Malho Rodrigues Founder & CEO

A few days ago, the owner of a small consulting firm told me his company had grown from 15 to 40 people in two years. Good news, right? Yes and no. Because the back office was still running the way it did when there were 15 of them. The HR person spent three days preparing each new hire. Customer support had queues of unanswered tickets. Sales spent 10 hours a week just booking meetings.

“I need to hire three more people just to manage the growth,” he told me. “But that means more fixed costs, more management, more complexity. And what if growth slows down?”

This is the classic dilemma facing Portuguese SMEs: growing by adding headcount is expensive and risky. Not growing means losing competitiveness. The third way — automating operations intelligently — has existed technologically for years. But only now has it become genuinely accessible to companies with 20, 50 or 100 people.

And here's the question I hear constantly: “Is this for me? Or is it only for large companies with technology budgets?”

The answer has nuances, so it isn't a simple yes or no. Those nuances are what I'll try to unpack in this article.

The dilemma is real: 42% of Portuguese companies point to the talent shortage as the main obstacle to growth. At the same time, margins are under pressure and the need to do more with less is constant. Hiring more isn't always the answer. Automating intelligently might be.

AI Agents and Agentic AI: from executing tasks to achieving business objectives

Let's start with an essential distinction that still causes a lot of confusion: the difference between the role of an AI Agent and an Agentic AI system.

An AI Agent is, in practice, a highly focused specialist. It performs a specific task and does it well: responding to customer requests, classifying emails, filling in forms, generating reports or scheduling meetings. The agent receives a task, carries it out within a clearly defined scope and returns the result. It's simple, predictable and effective.

Agentic AI is something else. It emerges when several of these agents work together, coordinated by a system that understands broader business objectives. This system doesn't just execute isolated tasks: it orchestrates agents, makes decisions in real time, adapts to context and pursues complex objectives. It can break a strategic goal down into subtasks, delegate them to specialised agents, track progress and adjust the plan when needed.

The practical difference is clear. An AI agent handles a support ticket. An agentic system redesigns the entire process: it prioritises requests by urgency and value, spots upsell opportunities, automatically updates the CRM, schedules follow-ups and measures customer satisfaction.

The agent solves a task; the agentic system achieves a business objective.

For many SMEs, this is genuinely transformative. Because the goal isn't just to reduce manual work, but to create operations that optimise themselves, make data-driven decisions and drive sustainable growth without having to triple the size of the team.

Where to start: the operations that make sense

There are three areas where I see an SME making almost immediate gains with AI Agents.

1. Internal operations Let's start with the most tedious, but also the most effective: extracting data from invoices. An agent can read hundreds of invoices a day and automatically extract VAT numbers, amounts and dates with accuracy levels above 95%, feeding the ERP or accounting software. In practice, this translates into a 25 to 50% reduction in the cost of repetitive administrative tasks. I've seen SMEs free up two or three people for higher-value work.

Another clear example is onboarding new employees. A set of orchestrated agents can create accounts in Google Workspace, Slack and internal systems, generate contracts for electronic signature, send welcome communications and schedule training. The result is 2 to 3 hours saved per new hire. In a growing SME, that scales quickly.

1. Sales and marketing Here the impact tends to be even more visible. A sales assistant on the website can qualify leads in real time, answer technical questions, book meetings and integrate directly with the CRM. In one specific case, a consultancy with around 20 to 30 employees saw clear results in three months: 40% more qualified meetings booked via the website, 15 fewer hours a week spent on manual scheduling and a 25% increase in the conversion of visits into sales opportunities.
2. Customer support Perhaps the area closest to genuine structural change. An agent doesn't just answer questions: it solves problems. It can reset passwords, process refunds, schedule technicians and interact with CRM, billing and internal databases. The numbers I see repeated are consistent: up to 80% reduction in resolution time, around 50% customer self-service and a significant increase in the satisfaction of internal teams, who no longer spend their days on repetitive tasks.

None of this is fiction. It's happening in Portugal today. But there's one essential condition for it all to work: quality data.

Data as a competitive asset (and why so many SMEs are flying blind)

In Portugal, 99.9% of the business landscape is made up of SMEs. These companies generate around two thirds of value added and more than 70% of employment. Even so, many struggle to grow and only a minority manage to scale sustainably. Why? The answer isn't a lack of ambition, some crisis or management whim. The problem is structural. Many operations are still designed around people rather than data. When everything depends on “I know, because I've been here for ten years”, automating becomes difficult, scaling becomes complex and applying AI becomes almost impossible.

Take, for example, a consulting SME with scattered data: proposals in Word, timesheets in spreadsheets, client history in Excel, conversations in Gmail and contracts spread across various folders in Dropbox. When a new client comes along and someone asks “how much did this type of project cost last time?”, the answer doesn't appear quickly, because the information is fragmented across different places and formats (when it exists at all). Now imagine trying to use an AI Agent to improve sales activity or forecast margins. The agent needs to read historical data, identify patterns and recommend prices. But it gets lost. Not for lack of intelligence, but because the data is fragmented.

Getting a company ready for AI therefore starts with organising the data. That means consolidating scattered information into an ERP, CRM or other consistent structures, defining key indicators aligned with business objectives, ensuring clear governance — who has access, who validates, how things are deleted — and guaranteeing security and GDPR compliance. Only then does it make sense to connect agents capable of reading, writing and acting on that information.

And there's funding available to support this path. The PRR's “Artificial Intelligence in SMEs” funding line covers up to 75% in non-repayable grants, up to a maximum of €300,000 per company, for projects integrating AI into processes, including data consultancy, software, equipment and hiring dedicated talent. This significantly lowers the barrier to entry and makes the decision to start far more accessible.

The infrastructure: how agents actually work

There's one technical detail worth understanding if you want to grasp how all this scales: the Model Context Protocol (MCP).

An AI Agent is only potentially useful. To generate real value, it needs access to data, tools and internal systems. An agent that can't read the CRM, query the product database, write an email or update a calendar may be intelligent, but it's powerless. This problem surfaced quickly across the industry: how do you let an agent access such different systems securely, in a standardised and scalable way? Every company uses different tools (Salesforce, HubSpot, custom CRMs) and each exposes different APIs. If every new agent and every new tool requires a bespoke integration, complexity explodes.

That's where MCP comes in. It's an open protocol created by Anthropic, often described as the “USB-C of AI”. MCP defines a standard way for agents to request data or perform actions in a company's systems. An MCP server exposes data and tools securely, while the agent, as an MCP client, consumes those capabilities through a common interface. In practice, this changes everything. When you decide to deploy a new agent, the path becomes much faster. Instead of integrating separately with the CRM, the ERP or the calendar, you simply describe what the agent needs to do and MCP handles the connections. The model is modular, scalable and secure, because it defines clear permissions from the outset.

It's no coincidence that OpenAI, Microsoft, Google and Anthropic have adopted MCP as a standard. In December 2024, Anthropic donated the protocol to the Agentic AI Foundation, under the Linux Foundation. This signals a clear industry convergence towards a model in which agents are tool-agnostic, as long as those tools expose an MCP server.

For SMEs, this has very concrete implications. When choosing an AI supplier or an integrator, you can and should ask: “do you support MCP?”. If the answer is yes, integration will be faster, safer and more scalable. If it's no, you'll be locked into point-to-point integrations that sooner or later become expensive and hard to maintain.

The supply side: Portuguese integrators and the market

Around these global “engines” (OpenAI, Microsoft, Google, Anthropic), consultants, agencies and integrators are emerging in Portugal that:

• Understand the reality of Portuguese SMEs (data maturity, limited technical know-how, budget);
• Adapt AI models to each company's reality;
• Help with compliance, security and integration with legacy systems;
• Offer vertical propositions by sector (AI for consulting, retail, industry, etc.).

Jelly works with Portuguese SMEs on this journey: diagnosing AI opportunities, running pilots, ensuring integration with internal systems and training in-house teams. This is the market starting to take shape.

The important point: you don't have to do this alone. And most SMEs can't. It's complex, it's new, it's risky. But there are partners available in Portugal who have already done this for other companies.

What SMEs need to do: from today to tomorrow

If you run an SME and you've read this thinking “OK, maybe this is for me,” what do you do from today?

First thing: diagnose.

Sit down with your team and look at your processes. Which work eats up the most hours, is repetitive and causes frustration? It might be customer support (hundreds of tickets a week), screening candidates (dozens of CVs) or processing invoices (hours of HR time each week). Identify 3 to 5 of those processes.

Then run a data audit. Where is your data? In centralised systems or scattered? What quality is it? Can you extract information reliably?

Second thing: start small.

Don't try to solve everything at once. Pick one process with clear impact. Ideally, something where you already have data. Start with a 2 to 3 month pilot, measuring real impact: time saved, errors reduced and satisfaction.

And this is where funding comes in. The PRR's “Artificial Intelligence in SMEs” line covers up to 75% in non-repayable grants for AI implementation projects, including consultancy, software, equipment, training and hiring up to 2 dedicated technical staff. This dramatically reduces the financial risk.

Third thing: think about integration and scalability from the design stage.

Don't deploy an agent in a silo. Ask yourself: how does this connect to the rest of the operation? Can the agent read the CRM? Can it write to the database? Can it call external APIs? If the answer is “no, it was all custom-built”, that will eventually be expensive to maintain.

So when you choose a supplier or integrator, ask about MCP, about integration standards and about scalability. It will save you thousands of euros down the line.

Fourth thing: build talent.

This is critical. You don't need 10 machine learning engineers, but you do need 1 or 2 people who understand: how to design processes for AI, how to write briefs for agents, how to validate results and how to supervise autonomous systems. This is hybrid talent: business people with technical understanding, or technical people with a business mindset.

And train your team. Talk about the opportunities in AI, about its limits and about the ethical questions. This reduces fear and increases responsible adoption.

Fifth thing: choose your partners well.

In most cases, you can't do this alone. You'll need partners: technology providers (such as OpenAI, Microsoft or Google), integrators who understand your reality and consultants with experience in data and implementation. Choose carefully. Go for partners who demonstrate practical experience, provide clear documentation and guarantee ongoing support over time.

Data, technology, people: the symphony of growth

Let me connect the dots.

There's a clear pattern among SMEs that manage to scale with AI. It isn't “we've got the coolest software”, but “we've got our data organised, we chose our technology carefully and we developed people in-house”.

This is a symphony, and every instrument needs to be in tune.

Data is the heart. Without quality data, no agent works. Technology is the instrument. OpenAI, Microsoft, Google and Anthropic are co-founding the Agentic AI Foundation to develop open standards, which means that in future there will be more interoperability, less lock-in and more freedom of choice. But people are the conductor. It's the business owner, the team or the project manager who asks: “How does this help our business? What risk are we willing to accept? How do we teach our clients? How do we protect quality?”

An SME that gets this right won't just be more efficient — it will be more resilient, more agile and, paradoxically, more human, because it frees people from repetitive tasks so they can focus on what really adds value: creativity, judgement and human relationships.

Final word

Ten years ago, everyone was talking about “digital transformation”. Many business owners didn't really understand what that meant. Some thought having a website was enough; others simply never did anything. Those who understood and acted are now well ahead. Now we talk about “artificial intelligence” and the pattern repeats itself. There's confusion, unrealistic expectations and fear too. Some think it's “ChatGPT for everything”, others would rather not touch the subject.

The truth, though, is simple: AI is a tool. And like any tool, the value you get from it depends on how it's used, with what intent and with what preparation. The Portuguese SMEs that understand this today, that organise their data, choose their technology well and build internal talent, will be the benchmark five years from now.

It's not about “being first”, but about being serious and wanting to do it well.

If you run an SME and you're ambitious about scaling, start today. With consultants, with partners, with specialist integrators. But start.

Gonçalo Malho Rodrigues works at the intersection of strategy, technology and creativity. Founder of Jelly Digital Agency, Stronddo and Scallent, he developed The Change Framework to help leaders rally teams around a cause.