Kelright

Seven questions to ask any platform that measures engineering in the agent era.

Engineering intelligence and developer productivity platforms read the systems a team already uses, such as GitHub, CI and a tracker, to report on how it delivers. They differ in what they count, where your data goes, and whether anything checks the work. These are the questions that tell them apart, and how the Kelright Benchmark answers each.

The questions

Ask each of these of every tool you consider, ours included.

Platforms such as Jellyfish, LinearB, Swarmia, DX and Faros are in this category. This page does not compare their features: those change often, and each is best read on that company’s own pages. It puts the questions that matter once agents write code, and says plainly what the Benchmark does about each.

What exactly is the before, and can it move?

Why it matters. An improvement is only a claim if the starting point cannot be quietly recomputed. If the baseline floats with the tool's rules, a dashboard can show progress that did not happen.

How the Kelright Benchmark answers. The Kelright Benchmark reads ninety days of history when you connect, freezes the before once every system has been read, and lets nobody recompute it afterwards, Kelright included. Every value is computed under a named version of the rules; when the rules change, the original window is measured again as a second before beside the first.

Before and after

Which of the numbers exist only because agents write the code?

Why it matters. The four DORA measures describe delivery whoever writes the code. They cannot say how much of the work agents do, what it costs, or whether agent work comes back more often than other work.

How the Kelright Benchmark answers. The Benchmark reads agent-assisted pull request share, AI spend per merged pull request, tests on agent-written pull requests, and cycle time and rework for agent-assisted work against other work, beside the four DORA measures.

The agent-native measures, defined

How does the tool decide which work an agent did?

Why it matters. Every agent-native number depends on attribution. A tool that does not say how it credits work to an agent gives you figures you cannot check, and a change in its method changes your trend.

How the Kelright Benchmark answers. The Benchmark credits a pull request to an agent from declared evidence: a co-author trailer, a declared agent account, a pull request label, or a join to the agent tool's own telemetry. The attribution rules are part of the named rule version, so a change in how credit is given is a new rule version, with the original window measured again under it.

Does your data, and your agents' telemetry, leave your systems?

Why it matters. Coding-agent telemetry shows what your team builds and how. Where it travels, and who can read it on the way, is a security decision before it is a productivity one.

How the Kelright Benchmark answers. The Benchmark reads GitHub and Linear read-only, by consent, and its measures arrive in your own Datadog under a key in your account that you can revoke. What your agents spend is read back from Datadog, where it already is, and nothing an agent emits passes through a Kelright server.

In your Datadog

Is any of the fee tied to the result?

Why it matters. A subscription is paid whether or not anything improved. Pay that depends on a measured change makes the vendor read the same before and after that you do.

How the Kelright Benchmark answers. For the Agentic Guild, part of the fee is paid at day ninety on the measured change from the baseline the Benchmark froze before the work started, and the rest is invoiced as each milestone's check passes. Kelright does not publish a price.

How the Guild is billed

Can a fund see its portfolio without seeing any company's private numbers?

Why it matters. A fund that wants one view across the companies it backs needs the companies to agree to be read. A view that exposes a company by default is one a company will not join.

How the Kelright Benchmark answers. The Benchmark's cohort view reads every company under the same rules and combines them by default, with no company named and no person among the data. A company's own numbers reach the fund only if that company opts in, in writing.

How a fund reads its portfolio

Does it check the work, or only count it?

Why it matters. A dashboard tells you how much agents did. It does not stop an agent's wrong change from merging, and it cannot tell you whether the change was what the company decided.

How the Kelright Benchmark answers. The Benchmark measures. The Guild installs the gates that hold agent-written code to your tests before it merges, and the Brain checks agent work against a cited record of what your company decided, holding a change that contradicts it even when every test passes. Each can be started on its own.

How the Brain checks agent work

When Kelright is not the fit

If you only want to watch activity, you may not need what we built.

The Benchmark is built around a before that cannot move and a check that closes every change. If what you want is a view of activity across many teams, with no frozen baseline and no gates, a tool built for that may suit you better. If you want to know whether agents made your team faster, and to hold their work to what you decided, start with where your team stands.

Benchmark your team

Opens an email to hello@kelright.com. Tell us where your code, your work and your Datadog live.