Skip to content

Does RPA still make sense? Coffee-time reflections on agents, skills and young people

RPA is not dying - it is shifting its place in the puzzle. Agents take over the thinking, robots stay for the doing, and the most interesting shift is happening in people. A personal, bird's-eye view of the market, teams and hiring.

Jacek Bugajski Jacek Bugajski ·

The industry has been announcing the death of RPA regularly for two years now - and the robots keep clicking away just as they always did. Invoices get posted, tickets get closed, audits pass. At the same time, the budgets, the conferences and the headlines already belong to AI agents. This gap between the narrative and the engine room intrigues me enough that I sat down with a coffee and wrote out what follows from it - for companies, for automation people, and for those who are right now wondering whom to hire. This is not an analyst report. These are coffee-time reflections - with claims you may well disagree with. And that is the whole point.

Three waves, one direction

From a bird’s-eye view, the automation of the last decade falls into three waves. The first is classic RPA: robots clicking according to rules, deterministic, predictable, excellent at repetitive work on structured data. The second wave added machine learning - robots began reading invoices, classifying documents, understanding messages. The third wave, the one we are living through now, is agents: systems that do not execute a script but reason, plan and pick the right tools for the goal.

Daniel Dines, the founder of UiPath, put it this way: robots gave us technology that mimics the way people do repetitive work, and agentic automation is, for the first time in history, starting to truly mimic the human mind. The scale of the market’s ambition shows in the statements of the past year and a half: Jensen Huang calls agents a new digital workforce and predicts that IT will become the HR department for agents, Marc Benioff talks about a multitrillion-dollar opportunity in digital labor, and Satya Nadella foresees the business logic of applications moving into agents.

Sounds like a death sentence for RPA? In my view, it is exactly the opposite.

Bots fail loudly, agents fail quietly

The best description of the difference between a bot and an agent that I have seen recently comes not from marketing materials but from a practitioner’s account. He replaced three classic bots with a single LLM-based agent. First week: euphoria, 80% of cases handled flawlessly. Second week: on edge cases the agent started making wrong decisions - and it made them with full conviction. It ended with adding guardrails and a human review queue.

The conclusion I have been repeating to clients for months: bots are reliable but dumb; agents are smart but unpredictable. Bots fail loudly and spectacularly - you see it immediately. Agents fail quietly - they do what they “think” you meant. In financial, HR or regulatory processes, a quiet failure is far more dangerous than a loud one.

That is why a mature agentic architecture does not throw the robots away. It assigns roles: agents think, robots execute, people lead. Deterministic execution is cheap, auditable and predictable - and that is exactly why it remains the foundation wherever a mistake carries a cost. We wrote about this at greater length in UiPath is no longer RPA and From RPA to digital workers.

The market is filtering itself anyway. Gartner forecasts that over 40% of agentic AI projects will be canceled by the end of 2027 - due to costs, unclear business value and a lack of risk control. Yet the same firm predicts that by 2028, 15% of everyday work decisions will be made autonomously. Both numbers are true at the same time: the hype will die, the technology will stay.

The skills landscape is shifting before our eyes

The most interesting change is not happening in the technology but in people. I observe it in my own company, at our partners’ and across hundreds of practitioner posts: the “RPA developer” role is turning, before our eyes, into an “agent engineer” role. And contrary to the panic visible in industry discussions, I believe this is a promotion, not a sentence.

An RPA developer does not, after all, start from zero. Three things they have done for years turn out to be an advantage in the world of agents. First, they think in flows, not scripts - branches, states and decision points are exactly the shape of an agent. Second, they have production discipline: retries, logging, monitoring, SLAs - a rare thing in the AI world, where many teams stop at an impressive demo. Third, they can turn vague business rules into reliable steps - and an agent only executes what someone designed wisely.

On top of that come new layers of competence: designing agents and prompts, integration with enterprise systems, orchestrating multiple agents and robots, process intelligence and - something I stress in every meeting - AI security. Guardrails, human control points, auditability of an agent’s decisions. On why writing the automation itself is today the easiest part, we wrote in A coding agent can write an automation in 10 minutes. Getting it into production takes 10 weeks.

And what about young people? Voices worth hearing

Here we reach the thread that moved me the most. If AI does more and more of the juniors’ work, why hire juniors at all? I hear that question more and more often - and the data show it is not theoretical. A study by Erik Brynjolfsson’s team at Stanford found that in the occupations most exposed to AI, employment of people aged 22-25 fell by a relative 16% since late 2022, while it kept rising for older groups. Dario Amodei, the head of Anthropic, openly warns about AI’s impact on entry-level white-collar roles.

And yet the leaders who are building this technology say, in unison, the opposite of panic. Matt Garman, the head of AWS, called replacing juniors with AI “the dumbest thing he has ever heard” - because juniors are the cheapest, the most engaged with AI tools, and a company that stops training them will, in ten years, have no one who can do anything. Daniel Dines says it plainly: giving up on hiring juniors is a mistake. Jensen Huang adds the perspective I consider the most accurate: AI will not take your job, but a person who uses AI will.

I subscribe to this in full. At SNOK we are not looking for people with ten years of experience in a single tool. We look for people who are open, curious and quick to learn - because the shelf of tools turns over every quarter, and curiosity does not age. A young person with a native instinct to reach for AI, plus an experienced architect who knows where automation hurts - that is the team for this decade.

Join the discussion

I have three questions I genuinely want your answers to. First: in your organizations, is classic RPA still doing the job, or is it merely waiting to be replaced? Second: how are you reskilling your automation teams - and what turned out to be the hardest part? Third: are you hiring juniors into automation teams today, or have you put it on hold “until things become clearer”?

Write in the comments on LinkedIn or directly to me. I will describe the most interesting threads in a follow-up piece - over the next coffee.

Sources

All quotes cited in the text come from public statements and publications; translations from English are faithful to the originals. Market forecasts are given with their publication date - they are forecasts, not accomplished facts.

The evolution of RPA and agentic automation

Market, numbers and forecasts

The labor market and young workers

Tematy: Agentic AI RPA UiPath skills recruitment
Found this useful? Please pass it on:

More from this series

Other

Weekly review: when AI starts to attack - agentic ransomware, a cascade of SAP CVEs and a turn towards sovereignty

The week the first fully agentic ransomware appeared, SAP froze costs to fund an AI push, and the market started saying the same thing: an AI agent is a new attack surface, and the answer is control - over the model, the data and the governance layer.

Weekly review: AI enters a phase of maturity - architecture, brakes and the first bills

The week AI got architecture, brakes and its first bills. SAP opens up a modular AI architecture with Claude as a component, UiPath adds Guardrails and case management, and the market starts talking about FinOps for AI and new attack vectors against agents that can see the screen.

NIS2 Autopilot: agent compliance under human oversight, powered by SecurityBridge

The NIS2 clock starts the moment an incident occurs: 24 hours for an early warning to the CSIRT, 72 hours for the full notification. The board is personally liable. NIS2 Autopilot combines SecurityBridge detection with an AI agent that triages the incident and drafts the report - while a human approves before anything leaves the organisation.

Get in touch

Get in touch Email office@snok.ai