Introducing Autonomous Workflows in Renewables Asset Management

Humanoid robots working in renewables won't have a body to climb turbines (at least not any time soon). But they'll work inside your workflows.

At BMW's Spartanburg plant in South Carolina, a humanoid robot called Figure 02 by Figure AI spent ten months working on a live production line. Five days a week, ten hours a shift, it picked up sheet metal parts and positioned them for welding. By the end of the pilot, it had moved more than 90,000 components and helped build over 30,000 BMW X3s. Its successor, Figure 03, has now started at the same plant, this time sequencing parts in logistics.

Robots in car factories are nothing new. Robot arms have been welding and painting car bodies for decades. So why does a humanoid feel like such a big deal?

Because the old robots and the new ones work in completely different ways.

A traditional robot arm is programmed. Engineers define every movement, every angle, every position and the arm repeats it inside a fenced cell. It is extraordinarily good at doing one thing exactly the same way, millions of times. Change the part and someone has to rewrite the program.

A humanoid is built for a different kind of job. It works in spaces designed for people, sees what's in front of it and works out how to handle it. Nobody programs every movement. It gets a task and it figures out the moves.

In the renewables industry, we don't expect to see humanoids climbing wind turbine towers any time soon. But that doesn't mean robots won't prevail in renewables asset management. The constraints are only true for physical robots. However, a robot doesn't need a body and most of the work in asset management doesn't happen on the turbine. It happens in asset performance management software such as Kavaken's platform, spreadsheets, reports, contracts and email threads. That is where a robot can go to work today, without needing legs and arms.

This is what autonomous agents bring to the renewables world.

Kavaken introduces autonomous workflows

KEN, Kavaken's AI agent, was already doing a lot. It answered questions about your portfolio. It ran analyses when you asked. It drafted the email to the OEM when Kavaken flagged an issue. Very useful, but now it can do even more.

KEN now runs your workflows autonomously. Instead of a question, KEN gets an objective and ownership of an operational outcome. It decides what to investigate next. It keeps track of where things stand over days and weeks. It reacts when new information arrives. It brings in human judgment when needed. And it doesn't stop until the job is done and verified.

Automation is not autonomy

"Autonomous" is one of the most overused words in tech right now, so it's worth being precise.

If every step is defined in advance, that is automation. It's the robot arm: fast, reliable and completely dependent on someone having foreseen every situation. Autonomy is different. You define the objective, the boundaries and the tools available. The agent decides what to do next based on what it finds.

An example: contractual availability reconciliation

If you manage wind assets under a long-term service agreement, you already know this one. Every period, the contractual availability is reconciled between you and the OEM. You check the availability report of the OEM against your own data (e.g., data from Kavaken's Outages module). Somewhere in the two sets of numbers, there are outages categorised differently: downtime the OEM counts as excluded, while you believe they should be deducted from the contractual availability. Each disagreement can have a price tag, particularly if bonuses and liquidated damages are in the picture.

The process itself has clear stages. Prepare your internal availability figures. Get the OEM's figures. Check against your own data. Define the gaps. Reconcile with the OEM. Reach agreement. Update the records. Close.

KEN triggers the process automatically when the month is completed and prepares the internal availability figures. It receives the availability calculations the OEM sent via email and compares them against the internal figures.

Say KEN finds four outages where the categorisation differs from the OEM's. It goes case by case and proposes the correct categorisation with the supporting justification:

Outage 1 is clear from the contract.

Outage 2 involves an ambiguous alarm code. KEN pulls the OEM's alarm definition before drawing a conclusion.

Outage 3 contradicts how similar outages were treated in previous periods. KEN references those cases to support its recommendation.

Outage 4 doesn't have enough evidence either way. KEN doesn't guess. It escalates to the asset manager with what it found and what is missing.

Now you have a well-prepared document prepared by KEN to reconcile the differences with the OEM. Upon your approval, KEN sends it to the OEM. Once the reconciliation with the OEM is complete, KEN updates the year-to-date figures and advises whether any bonus or liquidated damages are due.

Unlike automation, the process isn't linear, nor continuous. The system doesn't map all possible scenarios either. KEN is simply given the tools to achieve an objective while working within the defined guardrails.

Outcome tracking, not task tracking

Most workflow software tracks tasks. A report was received, a ticket was closed, a box was ticked.

An autonomous workflow tracks outcomes. The contractual availability reconciliation isn't completed until the disputed outages are resolved, the agreed position is recorded, and the contractual and commercial impact is calculated. Whether this is achieved in one email exchange or ten, KEN keeps the case open, follows up when the OEM responds and recalculates when new information changes the picture.

Why this isn't just another AI agent

Anyone can plug a general-purpose AI agent into their software today. The technology is widely available. What a general-purpose agent doesn't know is which nuances actually matter in renewable asset management processes, how an experienced asset manager works through them, what evidence holds up in a dispute with an OEM, when to escalate and what "done" really means.

That is where Kavaken's field expertise comes in. We teach KEN how an expert asset manager thinks. We also give it the tools to act: Kavaken's data analytics tools and machine learning models, contract and alarm data, and more than 1,500 field-verified cases. We bring the ingredients, equipment and recipes (including the secret sauces). KEN cooks.

Cover image by Alex Shuper from Unsplash