Executes
Completes whole tasks autonomously, without step-by-step supervision.
National Strategy · productivity and the AI economy
AI is not a
digital opportunity.
It is the only lever
on productivity.
Portugal has a twenty-five-year-old productivity problem and a business base made of micro-enterprises that cannot grow its way out of it. Artificial intelligence is the first technology capable of giving a five-person firm the capacity of a large one — without hiring.
01 · The problem
We are diverging,
not converging.
Labour productivity is measured as GDP generated per hour worked. It is the indicator that determines wages, tax revenue and capacity to invest.
€28,8
GDP generated per hour worked in Portugal, 2023
Eurostat / ENIA analysis71%
of the EU-27 average, which stands at around €71 an hour
Eurostat5.ª
lowest position in the European Union
Eurostat−6 p.p.
against 2000, when Portugal stood at 77% of the European average
ENIA analysisThe uncomfortable reading is not the position: it is the direction. While Portugal fell from 77% to 71% of the European average in two decades, economies that started further behind — Poland is the awkward example — converged. There is no sustainable incomes policy without reversing this curve, and no reversing this curve incrementally.
02 · Anatomy of the opportunity
The problem and the solution
are the same thing:
micro-enterprises.
96% of Portugal’s one and a half million companies have fewer than ten employees. The gap in value added per worker between them and large firms is enormous — and that gap is where the entire national recovery margin sits.
Instead of waiting for micro-enterprises to grow, we give them the capacity of large ones. That is an inversion of industrial policy, not a support programme.
96%
of Portuguese companies have fewer than 10 employees
INE · ENIA analysis€89k
of gross value added per worker in large companies
INE · ENIA analysis€35k
of GVA per worker in micro-enterprises — 39% of the large-company figure
INE · ENIA analysis9,4%
of companies with 10 to 49 employees use AI, against 49.1% of those with 250 or more
INE · IUTICE 202503 · The instrument
Generative AI answers.
Agentic AI executes.
The difference between a colleague who only answers when called, and a colleague who, given a task, does whatever it takes to finish it.
Plans
Takes an objective and breaks it into the necessary steps by itself.
Learns
Adjusts its behaviour according to the results it gets.
Orchestrates
Several agents complement each other, with outputs chained between them.
The leverage effect in a five-person micro-enterprise
42h
a week consumed by repetitive tasks: email, reports, answering frequently asked questions, invoicing and reconciliations
Before10h
a week after automating with agent orchestration
After+32h
freed each week — the equivalent of almost a full-time post, without hiring
GainENIA estimate drawing on McKinsey, Goldman Sachs (2024) and Atlassian (2025). It is a reference figure for sizing the opportunity, not a promised outcome: the real gain depends on each company’s task profile. That is precisely what the first 30 days of diagnosis are for.
04 · The National AI Agenda
Approved. Funded.
Now it has to happen.
Council of Ministers Resolution 2/2026 of 8 January, approving the National AI Agenda, its governance model and the 2026-2030 Action Plan.
32
initiatives across three pillars: Infrastructure and Data, Innovation and Adoption, Talent and Skills
RCM n.º 2/2026400 M€
of investment to 2030, mostly from European funds
Ministry for State Reform18-22 mM€
of estimated additional GDP, around 8% at peak impact
ANIA · fonte citada: McKinsey+2,7 p.p.
of additional productivity contribution to GDP growth over the next decade
ANIA · fonte citada: McKinseyWhat the Agenda gets right
It identifies the productivity deficit as the central problem rather than a side effect. It provides for compute capacity, an AI Factory and the national LLM AMÁLIA. It creates a Centre of Excellence within ARTE to avoid duplication across the State. It reinforces the “AI in SMEs” funding line and recognises that micro-enterprises need low-code products, not awareness campaigns.
Where ENIA presses
An approved agenda is the beginning, not the end — the test is delivery, and delivery is not published. There is no public tracking dashboard, no regulatory sandbox of the kind the AI Act requires by August 2027, and no involvement of those who have already adopted AI far above the European average: companies. An agenda designed entirely inside the State risks modernising the State and leaving the economy where it was.
05 · Recommendation for companies
A 90-day plan.
This is not a digital transformation programme. It is three months to get off the wrong side of the productivity curve — and future leadership starts with a plan.
Days 1 – 30
Diagnose
- Inventory every repetitive task the team performs more than three times a week.
- Quantify the weight of the routines: email, reports, customer service, invoicing.
- Identify the three processes with the greatest potential for immediate automation.
- Map the real cost of those tasks — hours × hourly cost — and prioritise.
Days 31 – 60
Pilot
- Deploy an agent on one simple, measurable process.
- Train the team and involve them in designing the agent, to head off internal resistance.
- Orchestrate using an automation and agent-coordination platform.
- Measure the time saved against the baseline and document the return.
Days 61 – 90
Scale
- Extend to two or three further processes, based on the pilot.
- Calculate return and break-even, counting internal effort.
- Set the twelve-month roadmap for agentic expansion.
- Communicate results and plans across the organisation and to partners.
Automating without classifying builds regulatory debt
Any agent that makes decisions about people — screening applications, assessing credit, prioritising patients — may fall under Annex III of the AI Act. Running the pilot and only then asking what risk tier it sits in is the most expensive way to find out. Classification happens alongside design, not after it.
06 · Public service
Get ready for the
AI economy.
Ask us how.
Beyond producing studies and surveys on the state of the nation, we believe a national AI entity has a duty to support those trying to succeed in this era. So we make specialists available, free of charge, to assess the best AI strategy for your organisation.
No cost, no sales pitch at the end and no compliance services on offer — that is a line we do not cross.