Frequently Asked Questions

Your AI return is how much of your AI tooling investment actually turns into shipped software. The formula is simple: (Outcome − Rework) ÷ Investment.

Most teams today measure only one side: how much code gets generated, how many licenses get used. But generating isn't shipping. A change written in twenty minutes that waits nine days for review, or bounces back twice over an ambiguous acceptance criterion, consumed AI budget without producing value.

It matters now because AI spend is already a visible budget line, and the question “is this paying off?” arrives before the evidence to answer it exists.

A DORA or engineering-intelligence dashboard tells you what's wrong. Leanmote does three things a dashboard doesn't.

First, it fixes the measurement before measuring. We map your board's workflow states so the metrics are true: it's common for states that actually behave as queues to be counted as active work, and that distorts everything downstream.

Second, we don't emit loose alerts. Every signal arrives classified into the pattern that explains it, paired with the action that fixes it.

Third, and this is the real difference: we put agents on executing that action. Our own installed base showed us the problem this solves: of all the signals an insight engine emits in a month, the vast majority are never acted on. Not from lack of interest: from lack of time. Detecting more doesn't move the needle; executing does.

A resolution agent is a unit of work, not a product feature. Each one owns a specific pattern in your flow: code review, rework, queues, deploy risk. And it executes the action that pattern calls for.

There are three kinds, and the distinction matters:

Resolution agents. They do the work: review a change, chase a frozen item, score a deploy's risk and generate its checklist.

Decision assistants. They don't execute: they turn a problem into a decision with an owner and a deadline, addressed to whoever has the authority to make it. Capping work in progress, for example, isn't something an agent can do: a leader decides it.

Foundation agents. They hold everything else up: they fix your board's state mapping and record what was recommended, what was actioned, and what moved the metric.

Being explicit about which is which is deliberate. Selling something as an “autonomous agent” when its lever sits in a human's hands is the fastest way to disappoint.

The general agent brings the method. Your operation brings the context, and that's what gets installed.

Concretely: the last 30 days of your workflows are the calibration set. The exact transitions where your work bounces back. The states that behave as queues on your board. The durations that are normal for your team and the ones that aren't. The projects where each pattern concentrates. A rework agent installed in your account looks for your transitions, not a handbook's.

And it keeps learning: every month of operation recalibrates thresholds and causes with the new evidence.

Agents earn autonomy through measured evidence, never on a schedule and never without your authorization. There are four levels:

Level 0 · Observe. Reads and records. Interrupts no one.

Level 1 · Propose. Presents the exact change with the evidence behind it.

Level 2 · Apply with approval. Prepares the change; a person confirms. Always reversible, with a dry run first and per-batch rollback.

Level 3 · Execute and audit. Acts on its own within a bounded domain, logging everything.

Every agent starts at the lowest level and moves up when it proves precision. And there's a hard limit: no agent reaches level 3 on decisions that affect people. That's product policy, not a technical limitation.

First things first: we don't store your source code or the content of your messages. Leanmote works on metadata: commit and pull request activity, work items. The diff never leaves your repository. And your code and your data are not used to train models, ours or anyone else's. It's written into the contract.

You connect over OAuth and authorize the scopes yourself. The baseline is read-only, all the diagnosis needs. Write access is granted by agent autonomy level and enables bounded actions: commenting on a review, moving an item. Never broad access. Every agent action is logged and reversible.

Underneath: TLS 1.2+ in transit and AES-256 at rest, infrastructure on AWS, and an independent SOC 2® audit by a licensed CPA firm. The full report is shared under NDA. Retention is defined in your contract, and on termination data is securely deleted.

Three sources: your board (Jira, Azure Boards), your repository (GitHub, GitLab, Azure Repos) and your CI/CD pipeline. Connecting takes hours, not weeks.

One honest caveat: coverage determines what we can offer you. We've seen accounts where we see the whole board but almost nothing of the code, and there the review agent isn't a candidate until the repository is connected.

That's why the diagnosis also reports what percentage of your process we're actually seeing, not just the findings. Before committing to dates, we validate access.

It's an analysis of your data, not a generic demo.

You connect your sources, and in the first session we work on what shows up: the patterns repeating across your teams, which boards each one concentrates in, the estimated impact on delivery value, and which agent would resolve each pattern, already calibrated with your signals.

That session also produces the baseline: the normal thresholds of your operation, against which improvements are measured later. That baseline is set with your numbers, not our estimates.

No, the opposite. Those tools do exactly what they promise: your team generates faster. That's not where the problem is.

The problem is that the capacity to generate grew and the capacity to review, integrate and deploy didn't. All that extra speed ends up in a review queue, in a ticket that bounces back, or in a deploy that breaks production. That's where your AI investment's return evaporates, and it never shows up in your AI tool's dashboard.

Leanmote works on that stretch. The more AI your team uses, the more valuable it is to repair the harness around it.

The first agent delivers in week 3: the board mapping fix, which is also what makes everything measured afterwards defensible. The code review agent, usually the highest-return one, lands around week 6.

The first measurement against baseline comes at 90 days. That window isn't arbitrary: it's the minimum for a flow change to be distinguishable from your operation's normal noise.

And something worth knowing up front: if your team doesn't act on anything, the record will show it. We'd rather the report say “recommended, not actioned” than have the renewal conversation be about perceptions.