We put a blunt question to developers on a recent live webinar: right now, could you actually prove AI agents are paying off for you or your team? Only 24% said yes. The other 76% were guessing, unsure, or already suspicious that agents are costing more than they’re saving.

That gap between adoption and proof is the real story in agentic development right now. Teams aren’t behind on running agents. They’re behind on knowing whether it’s working.

The proof gap

When we asked, “could you actually prove AI agents are paying off for you or your team,” the room split like this:

Right now, could you actually prove AI agents are paying off for you or your team?

Response	Count	Share
I have a gut feel, but no proof	37	45%
Yes, I’ve got the numbers	20	24%
Honestly, I can’t tell	18	21%
I suspect we’re spending more, not moving faster	7	8%

Add it up and 74% of the room is running on instinct, uncertainty, or a growing suspicion that agent work isn’t paying for itself. Not because the agents aren’t producing. Because nobody’s built the measurement layer to check.

What developers actually want to know

Everyone assumes the agent conversation is about cost. Ask developers what one number they’d want to see about their AI work tomorrow, and cost isn’t the answer most of them give.

If you could see one number tomorrow about your (or your team’s) AI work, which would you want most?

Response	Count	Share
Whether AI is efficient or needs babysitting	48	52%
What our models & skills actually cost	28	30%
Whether output is climbing or just spend	14	15%
How we compare to other teams / the org	1	1%

More than half want to know if the agent is actually working or just generating output someone has to babysit. Cost comes second. Team comparisons barely register. Developers don’t want a bill. They want to know if the thing is pulling its weight.

Multiple agents are already the norm

This isn’t a hypothetical for most of the room. We asked how many agents a typical developer on their team runs at the same time, and running more than one is already standard:

How many agents does a typical developer on your team run at the same time?

Response	Count	Share
One	37	43%
Two or three	33	38%
Zero, we’re not there yet	9	10%
Four or more	7	8%

90% of the room has at least one agent running. Nearly half are already coordinating two, three, or more at once. Coordination is the daily reality. Proof hasn’t caught up to it.

The subscription bill isn’t the biggest cost

When we asked what worries developers most about the cost of agents, token and subscription spend won by a wide margin. It’s the line item on an invoice, so it’s the easiest cost to see and the easiest one to worry about.

When you think about the cost of agents, which one worries you most?

Response	Count	Share
Token and subscription spend	36	45%
Rework from output that missed	22	27%
The review and merge bottleneck downstream	14	17%
Developer time spent supervising agents	8	10%

But rework from output that missed the mark and the review and merge bottleneck downstream add up to 44%, nearly matching the subscription line item. Those costs don’t show up on an invoice. They show up as a developer’s afternoon, a pull request sitting in review, a branch that has to be redone. The bill you can see gets budgeted. The cost you can’t see gets ignored, and it’s almost as large.

Where GitKraken fits

This is the gap GitKraken Insights and Kepler are built to close. Insights gives teams the number they actually asked for: whether AI work is efficient or needs babysitting, not just what it costs. Kepler is the delivery engine underneath it, built to coordinate agent-driven development directly:

• Multi-agent coordination across repositories, for teams already running two or three agents at once
• Commit Composer, for turning agent-generated changes into commits a human can actually review
• Conflict Resolver and automated rebasing, so agent output doesn’t stall in the merge queue
• Branch intelligence and merge-readiness insights, so a team can see what’s actually ready to ship, not just what an agent produced
• Human-in-the-loop workflows, so developers stay in control of what merges

The developer stays the one directing the work. GitKraken’s job is making sure that work is visible, measurable, and doesn’t pile up in review.

Kepler is available at gitkraken.com/kepler. GitKraken Insights is available at gitkraken.com/insights.

Questions from the room

A few questions came up live that our team is still following up on directly, including support for specific issue trackers and self-hosted environments, and how orchestration works across a top-level agent and multiple sub-agents on longer-running jobs. If you asked one of these during the webinar and haven’t heard back, reach out and we’ll get you an answer.

FAQ

Can most developers prove their AI agents are paying off?

No. In a live poll of webinar attendees, only 24% said they could prove it with numbers. 45% had a gut feel but no proof, 21% said they honestly couldn’t tell, and 8% suspected they were spending more without moving faster.

What do developers want to measure most about AI agents?

Efficiency, not cost. 52% said the number they’d most want to see is whether AI is being used efficiently versus needing babysitting. Only 30% pointed to raw model and token cost.

How many AI agents does the average developer run at once?

Among polled developers, 43% run one agent at a time and 38% run two or three simultaneously. Only 10% aren’t running agents yet.

What’s the biggest hidden cost of AI agents?

Rework from missed output and the downstream review and merge bottleneck, combined, worry nearly as many developers (44%) as token and subscription spend alone (45%). The visible cost is the bill. The hidden cost is the time spent fixing and reviewing what agents produce.

Is Kepler available now?

Yes. Kepler launched in Public Preview on June 16, 2026, as GitKraken’s Agentic Development Environment for coordinating AI agents across repositories, branches, and pull requests.

What’s the difference between GitKraken Insights and Kepler?

Kepler is where the agent work happens: task organization, commit structuring, conflict resolution, and merge-readiness. GitKraken Insights is the reporting layer on top, built to answer the efficiency and cost questions teams are asking about that work.