AI agents for engineering managers: more points from the same team, or you stop.
AI agents for engineering managers who need a number they can defend, not a pilot that quietly ends. You point us at one machine. We load the agents, attach them to your Jira board, and they start pulling stories off your backlog alongside everyone else. Nine sprints of trial and error are already behind us, so your team starts from the configuration rather than the experiment.
Claim a free beta place for my team
Free for beta participants, and beta participants keep the prices below for the life of the engagement.
The 5% and the 95%
MIT’s NANDA initiative found that 95% of enterprise GenAI pilots deliver no measurable impact on the P&L. The diagnosis was not model quality. It was brittle workflows and tools that sit to one side of how work actually moves.
The other 5% get exponential value. The difference is not the model they bought. It is whether the work flowing to the agent was shaped for it, which is the whole design of how we run AI agents for engineering managers.
An agent is only as productive as the stories it is handed. That makes this a Scrum problem before it is an AI problem, and it is why the backlog step is the one that decides your outcome. It is also the step almost every AI rollout skips, which is roughly the shape of that 95%.
What happened on our own team
Nine weekly sprints, June to August 2026, measured in Jira. Our agent worked the same board as everyone else. These are our numbers, not a vendor projection, and we wrote up the whole thing in our AI agent series.
Team throughput
+73%
44 to 76 completed points a sprint, first sprint to last, at the same human cost.
Agent vs human output
24 : 47
In the final sprint the agent completed 24 points to the team’s 47. Just over half of human output.
Share of assigned work
33.8%
Delivered by the agent by the final sprint, up from 7.69% in the first.
Agent-completed story points per sprint
Sprints 146 to 154 · weekly sprints · 16 June to 11 August 2026
3
5
4
9
9
12
18
18
24
The ramp is the cost of discovery: agent setup, Jira wiring, and learning which backlog patterns feed agents and which starve them. That is the part we packaged, so your team does not repeat those nine weeks.
Nobody on the team lost work: human-assigned output ended higher than it started, 36 points to 47. Treat +73% as first sprint to last rather than a steady weekly rate. Counts are assignment-based, so they measure work handed to the agent, not all AI assistance, and unassigned points sit outside both figures.
Source: our internal Human vs AI Delivery Report, Sprints 146 to 154, with Jira as the source of truth. What actually changed over those sprints, including what went wrong.
What you are actually buying
Not this
Nine sprints of your team’s time spent finding a working agent configuration, wiring up Jira, and discovering which backlog patterns feed agents and which starve them.
This
That configuration, deployed remotely onto your machine and your board, plus the backlog work that makes it produce anything at all.
How it works
Five steps. Four of them are ours. The operating loop we run is documented publicly, so none of this is a black box.
You point us at a machine
One box inside your environment. No procurement cycle, no new platform to license, nothing for your team to install.
We load the agents remotely
Setup is our work, not your team’s. Your developers stay on delivery while we do it.
We attach them to your Jira board
They work where the team already works. No parallel tool, no second place to check.
Agents start taking stories
Off the same backlog as everyone else, assigned the same way. Their output lands in your velocity data as AI points, so you can see exactly what came from where.
We work the backlog with your Scrum Master and Product Owner
Well-formed stories are what make agents productive. This is the step most AI efforts skip entirely, and it is the reason the numbers above happened at all.
What it costs after the beta
We price AI agents for engineering managers per developer per month. No setup fee, no minimum term, no platform licence.
Onshore team
$1,250
per developer, per month
For teams with a fully loaded developer cost of about $150,000 a year. Equivalent to $15,000 per developer per year.
Offshore team
$625
per developer, per month
For a fully loaded cost of about $75,000 a year. Same formula, same split. Mixed teams are priced developer by developer.
The price is set so that you keep at least half of what the agents save you. We assume a conservative 20% lift in points from agents working your backlog. At that lift your cost per point falls about 17%, and the price is half of that saving. Everything above 20% is yours, and the price does not rise as your gain does.
Worked example: five onshore developers
At the 20% assumption the price is built on
Fully loaded cost of five onshore developers
$750,000 / yr
Agents add 20% more points at the same cost
cost per point −16.7%
Value of those extra points at your old cost per point
$150,000 / yr
What you pay: 5 × $1,250 × 12
$75,000 / yr
What you keep at the 20% assumption
$75,000 / yr
Our own team reached agent output equal to just over half of human points by the final sprint. Every point past the 20% assumption widens the gap between what you pay and what you keep. The 20% figure is the basis for the price. Your own curve is the thing the beta is there to establish, on your board, so that the number you take into a review is yours rather than ours.
Stop at any sprint boundary
Billed month to month, per developer whose backlog the agents are working. When you stop, the agents and their configuration stay on your machine and keep running. What ends is our work with you, not your capability.
That is deliberate. If the number stops paying, we would rather you left than stayed out of contract obligation, because the only thing that makes this business work is teams choosing to continue.
Keeping the points honest
Any of this only means something if a point means the same thing in month six as in month one. Before the first sprint the agent builds a set of reference stories from your previous three months of Jira history, the sizes your team has actually been delivering.
Those references then apply to every story going forward, human-assigned and agent-assigned alike, and the system flags estimates that drift from them rather than quietly changing them. That protects you as much as it protects us. It is what lets you take an AI-points number into a leadership review and defend it when someone asks whether the points got smaller.
The resistance is the real work
The hard part of this is rarely technical. It is a developer who has decided the agent is a threat, and who is right to want an answer before they cooperate.
Three things move someone from resistance to action. Attitude: an attitude of resistance keeps a person stuck, and the useful question is why it is there rather than how to override it. Motivation: people move for what matters to them personally, not for a throughput target on your slide. Responsibility: the duty to build something better for yourself, your family and the people you work with.
Developers who engage with this early become the ones who can put agents on a team and show the gain in the board’s own data. That is a scarce and well-paid skill. Developers who wait become the people who resisted the last change, and every industry has them. This is a challenge to change, and it is worth making it directly to your team rather than around them.
What happens after you sign up
Three steps, and none of them is a sales call.
We reply to you personally
A real person from the team, not an automated sequence. We will ask a few questions about your board and your backlog.
A short call to check fit
Thirty minutes. If your backlog is not ready for agents, we will say so and tell you what would need to change. That is useful either way.
We schedule your first sprint
You point us at the machine, we load the agents and attach them to your board. Your team keeps delivering while we do it.
Prove it in your own organisation
The question you actually have
The figures here come from one team, ours, over nine sprints. The question you have is not whether it worked for us. It is whether it works on your board, with your backlog, inside your organisation, because most things that work elsewhere do not survive that test.
The beta is built to answer exactly that question, on your own board and in your own Jira data, at no cost. We are not asking you to trust our curve. We are giving you the means to produce your own, and the first teams through are the ones whose results the rest of the market reads.
Who these AI agents for engineering managers suit
Teams already running Scrum with a real backlog in Jira. If the backlog is thin or the board is decorative, the agents will have nothing to pull and this will not work.
Engineering managers who can get one machine stood up in their environment without a six-month approval cycle. During the beta that speed matters more than the size of your organisation.
Questions
What does the beta cost?
Nothing. Beta participants take part free. The beta exists to prove the system inside real organisations, on their own boards, and your results are what let us do that for the teams who come after you. That is worth more to us right now than a fee.
Beta participants also keep the prices on this page for the life of the engagement, whatever we settle on for everyone else afterwards.
What if my team does not hit 20%?
Then you will know it from your own Jira data rather than from a slide, and you stop at the next sprint boundary. Either way you leave with an answer about your organisation, which is the thing no one else’s case study can give you.
What happens to the agents if we stop?
They stay. The agents and their configuration are deployed on your machine and attached to your board, so when the engagement ends they keep running. What ends is our work with you, not your capability.
Will this replace developers?
It did not on our own team. Human-assigned story points ended the nine sprints higher than they started while the agent share of assigned work rose from 7.69% to 33.8%. Human output did not fall as the agent took on more.
What we cannot honestly claim from one team is that the same holds everywhere, which is part of what the beta is for.
How much of my team’s time does this take?
Setup is ours. The backlog work in step five needs your Scrum Master and Product Owner, because it is a conversation about your stories and nobody outside your team can have it for you.
What is not included?
The machine the agents run on, and the model and compute they consume. Those stay on your accounts deliberately, so you see the true cost per point rather than a number we have marked up. We will show you what ours runs to.
Claim a free beta place for your team
We are taking a limited number of teams into the free beta for AI agents for engineering managers. We are looking for managers running real Scrum on a real backlog in Jira. Leave your details and a real person will get back to you. No commitment, no cost, and if it is not a fit for your backlog we will tell you that too.