UiPath and Warmwind: Agentic Automation’s Two Markets

14 min read

TLDR Tap for the short version

Klarna famously cut 700 customer service jobs for AI back in 2024, only to quietly bring humans back once quality took a hit. That same shift toward a hybrid model is now hitting a different part of the tech world: back-office software. Here, enterprise giant UiPath and a tiny 15-person German startup called Warmwind are placing two completely different bets.
Both are using AI agents that can actually reason through a task instead of just following a rigid script, but they’re pitching them to totally different customers. Warmwind is going after the small businesses that big tech vendors usually ignore.
UiPath sees a hybrid future where agentic automation handles unpredictable work while traditional scripts cover the routine stuff, all managed in one place. Warmwind, on the other hand, is selling the exact same tech as straightforward, self-serve labor.
You can see the difference in how they charge: UiPath bills by the transaction, while Warmwind advertises a flat hourly rate.

A lesson Klarna already learned

In February 2024, Swedish fintech Klarna’s chief executive announced that an AI assistant, built with OpenAI, had done the work of 700 customer service agents in its first month, handling 2.3 million conversations and cutting resolution times from eleven minutes to under two. The tech press treated it as proof that AI could replace entire departments. Fourteen months later, Klarna was rehiring humans. CEO Sebastian Siemiatkowski admitted that cost had been weighted too heavily as a success metric, and what the company got back was lower quality. By 2026, Klarna had settled on a hybrid model where AI handles the routine questions, humans handle disputes, hardship cases and anything that needs judgment.

We are seeing that exact same mindset pop up in another part of the tech world. It is not customer service this time, but the behind-the-scenes work like processing invoices, shifting data around, and doing all the tedious screen-clicking a human used to do manually. Just last month, a small startup in Germany called eva AG laid out their exact prices for an AI worker.

You can rent one for just a euro or two an hour, or pay a couple of cents every time it processes an invoice or finds a sales lead. Their tool, Warmwind, works by reading the screen like a person would, figuring out which number is the actual total, and putting it where it needs to go, even without a preset layout. Old-school automation bots had to be taught every detail ahead of time, and the second a layout changed, everything would break.

Just eight days later, UiPath put out its latest quarterly results. The company specialises in traditional RPA and agentic AI, and while it actually beat Wall Street’s revenue estimates, its stock still took a hit. On paper, the numbers looked great. UiPath reported $410 million in revenue for its second quarter of fiscal 2027, which was up 13% from the year before. Annual recurring revenue climbed to nearly two billion dollars, and its retention rates were sitting steady at 109%. On top of that, eighteen of their top twenty deals included an AI feature.

Even so, the market barely blinked. It is worth keeping in mind that those numbers mostly come from selling high-end automation to deep-pocketed enterprise clients who already have the tech expertise to run it.

Warmwind, on the other hand, is selling to customers without an existing automation platform to upgrade. Its published pricing is built for a buyer discovering automation for the first time. Both companies are taking the same underlying capability, an AI that reads and decides instead of following a script, into two very different parts of the market. And like Klarna, both are wagering on how much of that judgment to leave with the AI and how much to keep with a script, an orchestration layer or a person.

RPA was built for certainty

Over the past twenty years, automation was a completely different story. The earliest and still most common form is rule based robotic process automation, where software follows a fixed script against a known screen. Click a field, copy a number, paste it somewhere else. It’s reliable because it’s narrow. It’s also brittle. One change to the layout it was built for, and it breaks until someone rewrites it.

Two other ideas grew out of that main approach. API integration lets systems talk to each other directly instead of a robot clicking around on a screen. It is faster and more reliable, but it only works if that feature is built in. Visual automation works with almost any software by watching the screen and clicking around like a real person. The catch is that getting a computer to actually read a screen correctly is a much harder problem.

AI models now extend that visual approach further, which is the shift the Warmwind invoice example shows in practice. Instead of a script telling the software what to click and where, an AI model looks at the screen, reasons about what needs to happen, and decides its own path through the task. This is agentic automation. It’s the step from software that follows instructions to software that interprets and decides, and it’s what both companies in this piece are building around.

UiPath isn’t killing the bot

UiPath is the clearer of the two cases to trace, because it has left the most paper behind. Its own public material doesn’t describe agents replacing RPA. It describes agentic automation as the next stage after RPA and AI powered automation, with agents working alongside deterministic robots rather than instead of them. The company’s 2026 annual report frames its evolution as a move from UI and RPA automation toward a platform that combines automation, AI agents and people.

Simple, predictable work is still best handled by traditional automation scripts rather than AI reasoning models. AI agents really show their value on the unpredictable edges, like reading an unexpected document format, picking the right next step, or fixing an error that would freeze an old script. UiPath is currently building its tech around managing agents, classic robots, and humans together. It is a clear bet that big companies will land on a hybrid approach instead of going fully AI, which is the exact same lesson Klarna learned the hard way.

What actually gets automated is changing, too. RPA used to tackle just one specific task at a time. Today, an orchestration layer can map out the whole process from start to finish, choosing whether a robot, an AI agent, an API, or a real person handles each step. You can see this exact shift in UiPath’s Maestro documentation, which shows RPA tasks, AI agents, API calls, and human sign-offs all running together in a single flow.

Companies selling governed platforms, deep integrations, audit trails and security controls are adding agents as a new capability inside a structure their customers already trust. The buyer is a large organisation with a procurement process and an existing automation program to expand, not one starting from nothing.

Warmwind starts where enterprise RPA never did

Warmwind goes after smaller businesses that lack all that heavy infrastructure. A small company usually has zero procurement process and no patience for a long enterprise sales pitch. Warmwind pitch its product as autonomous AI workers that run software visually on isolated cloud computers. It uses the exact same tech UiPath relies on, just wrapped up for a buyer the big automation giants never bothered to go after. It is sold more like hiring help than buying software, priced simple enough to get without a sales call, and built to be totally self-serve rather than negotiated.

Automation has been sold four different ways as it’s matured. A licence or platform capacity, then consumption such as execution time, then a completed transaction like a processed invoice, then an agreed business outcome. Each step moves more of the performance risk from buyer to seller. A customer on a licence eats the cost of a bot running at half efficiency. A customer paying per completed invoice makes retries and failures the seller’s problem.

UiPath has moved further along that ladder than a testing phase. Its documentation now specifies transactional billing for eligible agentic Maestro processes, including how a retry consumes AI Steps within a transaction. The meter applies to a whole agentic solution run across products rather than to Maestro itself, another sign that the commercial unit is moving from an individual bot or component toward the orchestrated process. Outcome based pricing, where the customer pays only for a result the seller stands behind, remains the direction UiPath’s founder and CEO, Daniel Dines, described to analysts on the September earnings call as still emerging, on top of its existing licensing. Warmwind has published a flat hourly rate from the outset.

Warmwind’s flat hourly rate is designed to be understood without a sales call. UiPath’s transactional billing ties enterprise usage more closely to cost. Neither company has published what a successfully completed task costs once retries and corrections are counted, which limits how far this comparison can be pushed. What each has chosen to charge for still tells you who each company thinks its buyer is.

Orchestration is where the value may sit

Pricing alone will not determine which suppliers capture the market. As agentic systems take on more of a workflow, something still has to decide which step goes to a script, which goes to a model, and which goes to a person for approval. That coordination function, orchestration, is a plausible place for value to concentrate as the technology matures. It’s the same governance question IC covered in Europe’s new market for shadow AI monitoring. Owning it doesn’t require building the single most capable agent. It requires being trusted to govern many of them, including agents a company didn’t build itself.

Companies already running deterministic automation at scale, UiPath among them, have a real case for owning that layer, though it’s unproven. It depends on whether enterprise buyers end up wanting one trusted control layer, or a looser mix of best of breed agents from different suppliers. UiPath and Warmwind aren’t the only ones making bets here. Microsoft, Salesforce, ServiceNow and Automation Anywhere are all building comparable agentic capability into adjacent categories, some extending existing enterprise footprints and others reaching toward smaller buyers, in patterns similar to what IC has documented in how Rezolve AI leans on partners rather than direct sales.

On the downmarket route, reliability and supervision matter as much as price. The self-serve model’s core assumption is that cost and technical complexity are what’s kept small businesses out of automation. The most recent OECD survey of small and medium enterprises, covering more than 2,000 businesses across 12 countries between the fourth quarter of 2025 and the first quarter of 2026, points somewhere else.

Up to 70.9% of EU businesses that considered adopting AI but didn’t cited a lack of relevant skills or expertise, ahead of cost. 64% said they need direct financial support to advance their digitalisation efforts at all. A published per task price removes a cost barrier. It doesn’t obviously remove a capacity one, and public material describes products like Warmwind’s as self-serve without establishing an onboarding or implementation program beyond the product itself.

Regulation sits closer to a supervision cost than a barrier for this segment. Warmwind’s positioning around German infrastructure and GDPR compliance targets SME buyers wary of routing workflow data through a US based provider, though data residency alone doesn’t establish that no US based subprocessor ever touches that data. The EU AI Act’s high risk obligations, the tier covering credit scoring or employment screening, were pushed back to December 2027 in a May 2026 provisional agreement, and invoice processing or customer support automation generally sit outside that category regardless. GDPR, not the AI Act’s high risk tier, is the more immediate compliance question for most of what these tools do.

Life after RPA

RPA is still great for predictable work, but it is quickly becoming just one piece of a much bigger automation puzzle. AI agents bring interpretation and flexibility, traditional systems handle the steady consistency, and orchestration software ties them all together with human sign-offs. That is the exact balance Klarna eventually built by hand after its all-in bet on AI fell short.

Established tech platforms could easily evolve into those central control layers, blending agents, fixed scripts, compliance, and human approvals into one package big corporate buyers prefer over standalone tools. On the flip side, agents might simply turn basic automation into a cheap commodity, making simple web and office tasks affordable enough for small businesses to jump on, while putting serious pressure on traditional software licenses. The market could also just split by risk. Low-stakes jobs like research and lead finding might go fully automated, while high-stakes tasks like processing payments stay tightly supervised in a hybrid setup.

At the end of the day, reliability, supervision, workforce skills, and the true cost of mistakes will dictate what works best for each business. UiPath is expanding its enterprise platform, while Warmwind is testing whether that same core capability, an AI that reads a screen instead of following a rigid script, can be sold directly to smaller companies that vendors used to ignore.

Watch how fast smaller businesses actually adopt this, what a completed task costs once retries and corrections are counted, and how much vendor revenue shifts from licences to pay-per-use and outcome pricing. UiPath’s own outcome based model, still emerging by its CEO’s own account, is one concrete thing to track over the next few quarters.

Methodology note. Company-reported figures, including Warmwind’s registration and revenue figures, are identified as self-reported and unaudited where used. Klarna’s 2024 figures come from the company’s own announcement; the 2025 reversal is drawn from Klarna executives’ public statements to Bloomberg and CX Dive. No independent, task-level completion-rate or reliability data for Warmwind’s or UiPath’s agentic products was found in public sources.

Spotted an error? Contact [email protected].

FAQ

What’s the difference between traditional automation and agentic automation?

RPA follows a fixed script against a known screen or interface. Agentic automation adds an AI model that looks at the screen, reasons about the task, and decides its own path through it, rather than following pre-written steps.

Is UiPath being replaced by AI agents?

UiPath’s own material doesn’t describe that outcome. It positions agents as handling unpredictable work alongside deterministic automation, with orchestration software coordinating both, rather than agents replacing the scripts that already work well.

Does the EU AI Act cover invoice processing and customer support automation?

Mostly not under the high risk tier, which mainly covers use cases like credit scoring or employment screening and whose deadline was pushed to December 2027. Transparency rules and general purpose AI model obligations already apply regardless, alongside GDPR.

What would show which route is winning?

Adoption among the small businesses enterprise automation never reached, the real cost of a completed task once corrections are counted, and how much of the industry’s revenue keeps shifting from licences toward usage and outcome pricing.

UiPath, a 15-Person Startup and the Emergence of Automation-as-Labour

Contents

A lesson Klarna already learned

RPA was built for certainty

UiPath isn’t killing the bot

Warmwind starts where enterprise RPA never did

Orchestration is where the value may sit

Life after RPA

FAQ

Like this:

UiPath, a 15-Person Startup and the Emergence of Automation-as-Labour

Contents

A lesson Klarna already learned

RPA was built for certainty

UiPath isn’t killing the bot

Warmwind starts where enterprise RPA never did

Orchestration is where the value may sit

Life after RPA

FAQ

Share this:

Like this:

Discover more from Industry Contents