The Signal #118 — Dakota’s read on the AI news that actually matters to people running a business.

The pitch was simple. AI handles the boring parts, you get your time back, everyone goes home earlier.

That pitch has not matched reality for a lot of people running operations right now. And one engineer decided to stop guessing about it and look at the actual numbers.

What happened

Orestis Ioannou, VP of Technology at Bold.org, wrote a post this week doing something most people in this conversation do not bother with. He pulled his full GitHub contribution history from January 2019 through October 2026 and mapped what AI agents actually did to his workload over time.

The data is worth walking through slowly.

His busiest coding years at Bold.org were the early ones. His highest year was 2020, with 2,711 contributions. That number declined steadily as his role grew. By 2023, his quietest year, he logged 1,189 contributions. His lowest single month since joining Bold.org was December 2025, with 50 contributions.

Then 2026 happened. By October 27th of this year he was already at 2,102 contributions, more than any full year from 2021 through 2025. August 2026 was his single busiest month in the entire timeline: 427 contributions, against a previous high of 319 in April 2020. His median weekday went from 3 or 4 contributions per year across 2022 to 2025, to 7 in 2026. He went from 40 to 46 weekdays per year with zero GitHub activity down to 18 through October 27th. From 2022 to 2025 he had exactly one day per year with 24 or more contributions. In 2026 he already has 16.

Here is the part that makes this interesting. In 2020 when those numbers were similar, he was a full-stack engineer. Writing code was his entire job. In 2026 he leads engineering. On paper his GitHub should look like his quietest years. Instead it looks like he never stopped being a full-stack engineer.

His explanation is direct. Agents did not make his existing workload faster to finish. They made it easy to give himself a much bigger one.

Why it matters

This is not a complaint post, and it should not be read as a warning to avoid AI tools. Ioannou is explicit that he can now investigate and ship things that would not have been worth starting two years ago. The tools are working.

The operational insight is different. When the cost of starting something drops, the number of things you start goes up. That is not a surprise in theory. But the shape of it in practice catches a lot of operators off guard.

He describes his old week as two or three serious tasks. One problem in his head at a time, finish it, move on. Today he runs several agents in parallel: one implementing a change, one investigating a production question, one working through a review, another doing research he will need later. None of them needs him constantly. Every one of them needs him sometimes.

That word, sometimes, is where the friction lives. Each agent eventually surfaces a question, a plan to approve, a diff to read, a result to interpret. Each response requires switching contexts, reloading a different problem, a different set of decisions. Two or three serious tasks was tiring in a familiar way. Managing many agents is tiring in a new way.

If you are running a team that has adopted AI tools, or thinking about doing so, this is the thing to track. Not just output volume, but where human attention is actually going after the tools go in.

What most people get wrong

The common assumption is that AI creates capacity by compressing time. You finish the same amount of work faster, and the freed hours go somewhere useful.

What Ioannou’s data shows is a different mechanism. The ceiling on what feels worth attempting drops. A bug that was not worth fixing before now is. A pattern that needed cleaning up but was never a priority now gets a pull request. A project that would have sat on a list for a year now becomes a sprint.

The capacity does not become free time. It becomes more work, usually work that was previously invisible because it was never considered possible.

For an operator in e-commerce, that might mean your merchandising team suddenly has the ability to test twice as many campaign variations, and now someone needs to review all of them. For a real estate brokerage, it might mean your agents can generate property analyses in minutes, and now the volume of analyses in flight at any given moment has tripled. The bottleneck does not disappear. It moves.

Ioannou also notes the pattern of bursts followed by wind-downs in his monthly data. January and February 2026 were his first big burst as he started working seriously with agents. Activity pulled back through the spring. Then April spiked again, partly because model improvements renewed his energy. This is a rhythm worth anticipating, high engagement, then a quieter period of processing and recovery, not a steady linear climb.

The lesson operators can take from this

AI tools expand the surface area of what your team will attempt. That is mostly good news. But expanding what people attempt without adjusting how attention is managed is a setup for a different kind of exhaustion than anyone planned for.

The question to ask is not just what can we do faster now. It is what are we choosing to take on that we never would have before, and do we have the decision-making capacity to handle what comes back from all of it at once.

That is a management question as much as a technology question.

If you are working through what AI adoption actually changes about how your team operates, not just what it produces, the thinking at xovionlabs.com is a good place to keep reading.