What engineering leaders, funds and operators ask before they start.
Answers on agentic engineering and on Kelright's three offerings: what each does, what it needs from you, how it is billed, and who sees what. Each answer stands on its own, so you can link to the one you need.
Prefer a link by email?hello@kelright.com
About Kelright
What Kelright is, who it is for, and what it does not claim.
What is Kelright?
Kelright helps software companies get the most from AI coding agents. It has three offerings that run as one loop: the Agentic Guild installs agentic engineering practice in a team's repository and CI, the Benchmark proves in the team's own numbers whether it worked, and the Brain keeps a cited record of what the company decided and checks agent work against it. Each can be started on its own.
How the loop fits togetherWhat is agentic engineering?
Agentic engineering is building software with AI coding agents that take on real work, such as writing, testing and reviewing changes, while the team's own tests, gates and review rules decide what merges. It differs from using an assistant in the editor because the checks go up before the agents do. Kelright's Agentic Guild installs the practice in a team's repository and CI.
The definition, with related termsWho is Kelright for?
Three readers. Engineering leaders at software companies with roughly eight to sixty engineers, whose teams write code with agents already or are about to, use the Guild and the Benchmark. Venture funds use the Benchmark to read the companies they back as a cohort. Operators, such as a CEO, COO or head of product or operations, use the Brain to make sure what their engineers and their agents build is what the company decided.
Who is behind Kelright?
We have done this work three times: five years forward-deployed at Palantir, where we founded its Pipelines Guild; running engineering at a startup built with coding agents before the practice had a name; and the turnaround of a global franchise business more than fifty years old, from the CTO's chair. We also hold three granted US patents.
The record, by who it matters toDo I have to buy the Guild, the Benchmark and the Brain together?
No. The Guild, the Benchmark and the Brain can each be started on its own. They are designed as one loop: the Guild changes how a team builds, the Benchmark proves in the team's own numbers that it worked, and the Brain keeps what gets built to what the company decided.
Are the charts on kelright.com real data?
No. Every chart and sample on this site is labelled Illustrative or Sample read, and its positions are made up to show the shape of the thing, not measured from anyone. Kelright publishes a measured number only with the system it came from and the date it was read. The record of who is behind Kelright is not illustrative: it is what we have done, with links where it can be checked.
The Agentic Guild
Thirteen weeks, a fixed scope, and a check that closes every milestone.
How long does the Agentic Guild take?
The Agentic Guild runs for thirteen weeks at a fixed scope. It has seven milestones, M0 to M6, that overlap as the work does, and each is accepted by a check in your own systems. The install starts with a deposit when the contract is signed and ends with a handoff in the last two weeks.
Each milestone and its checkWhat are the Guild's milestones?
The Guild's milestones are M0 Diagnostic (weeks 0 to 2); M1 Instrumentation and cadence (weeks 1 to 3); M2 Test harness (weeks 3 to 6); M3 CI and delivery (weeks 5 to 9); M4 Tech-debt tranche (weeks 7 to 11); M5 Agentic workflow adopted (weeks 8 to 12); M6 Handoff (weeks 12 to 13). The weeks overlap because the work does. The milestones are the same for every team; the targets each one is accepted against are set from where the team starts.
What each milestone deliversWhat does it mean that a Guild milestone is accepted by a check in my own systems?
A Guild milestone is done when something in your own systems says so: CI reports the coverage, the pipeline reports its time, the tracker shows the debt register closed, or the Benchmark reads the share of the work agents did. You can watch each milestone hold without relying on Kelright's word.
How is the Agentic Guild billed?
The Guild is billed as the checks pass. There is a deposit when the contract is signed, then two invoices that each wait for their milestones' checks: one when M2 and M3 are accepted, one when M4 and M5 are accepted. At day ninety, part of the fee is paid on the measured change from the baseline the Benchmark froze before the work started. The fee is fixed before work begins and set by the size of the team. Kelright does not publish a price; ask for a quote.
How the install is billedWhat access does the Guild need to my code?
The Agentic Guild needs read access to your repositories, CI and tracker. Every change Kelright makes arrives as a pull request through your own review.
How much of my team's time does the Guild take?
The Agentic Guild needs a lead who owns the install with Kelright, two engineers who become its champions, and a few hours of the team's time in the diagnostic.
What is left running after the Guild finishes?
The practice stays in your repository when the Guild hands off. The guardrails are agent instruction files that carry your conventions, CI gates that hold a coverage floor, require tests on agent-written changes and keep changes small, and branch protection and review rules to match. The Benchmark keeps reading your delivery against the frozen baseline. The distance to the best-performing teams on each measure is cut into issues sized for one engineer or one agent. Two of your engineers are trained as champions who run and extend the guardrails.
What staysThe Benchmark
How an engineering team's delivery is read, and what a fund sees.
What does the Kelright Benchmark measure?
The Kelright Benchmark computes 6 families of measures from the systems your team already runs: Delivery, Review flow, Work flow, Agent adoption, Agent economics, Quality. Delivery is the four DORA measures. Agent adoption and Agent economics only exist once agents write code. Every value says which system it came from and is computed under a named version of the rules.
Every measure, by familyWhat are agent-native engineering measures?
Agent-native measures are the ones that only exist once AI agents write code. The Kelright Benchmark reads how much of the work agents do, such as agent-assisted pull request share, and what it costs, such as AI spend per merged pull request. It also compares cycle time and rework for agent-assisted work and other work, so a team can see whether agents are making it faster or only making it spend more.
Definitions of the agent-native measuresWhat is a baseline frozen once, and why does it matter?
When you connect, the Kelright Benchmark reads ninety days of history from each system so there is a before on the first day. The before is frozen once every system you connected has been read, and nobody can recompute it afterwards, Kelright included. Every later reading is made under the same named version of the rules. If the rules change, the original window is measured again under the new rules as a second before beside the first, and the first is never rewritten. Better than before is only a claim if the before cannot move.
Before and afterWhich systems does the Benchmark read?
The Kelright Benchmark reads GitHub for pull requests, reviews and deployments, Linear for the work and how it moves, and your own Datadog, where it reads back what Datadog already measures, including what your coding agents cost. Access is read-only, and only by your consent.
Does what my coding agents emit pass through Kelright's servers?
No. The Benchmark's measures land in your own Datadog, under a key in your account that you can revoke. Agent spend is read from Datadog, where it already is, and nothing an agent emits passes through a Kelright server.
In your DatadogHow does a venture fund see whether its portfolio is getting faster with AI?
A fund reads its cohort as one: every company under the same rules, quarter after quarter, combined 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. The Guild can also run across the companies a fund backs as a cohort, and each company decides for itself.
How a fund reads its portfolioCan the Benchmark compare my team with other teams?
Yes, by design. Every team is read under the same rules, so where your team sits on each measure can be read against teams that work with agents. Peers are combined, no team is named, and your numbers are never shown to anyone you have not chosen.
Against peersThe Brain
How an agent's work is held to what the company decided.
What is the Kelright Brain?
The Kelright Brain turns your meetings and documents into a cited record of what your company decided, with your people approving whatever the Brain is unsure of, and holds the work AI agents do to that record. Code is checked against its tests; the rest of what agents do is checked against the Brain.
How the Brain keeps agents correctHow does the Brain check what AI agents do?
The Brain works in four steps with a person at the two that decide what the record says. It captures decisions, commitments and owners from your meetings and documents, each with the sentence it came from. Your people approve what the Brain is unsure of. Agent work is then checked against the record before it acts, and a change that contradicts a decision is held, with the decision named. If the work was right and the record is out of date, the Brain proposes the change and a person approves it.
Does the Brain replace tests or code review?
No. The tests ask whether the code works; the Brain asks whether it is what the company decided. A change can pass every test and still be held, because a decision your people approved says it needs, for example, a second reviewer. The Brain's check sits beside the tests.
What does the Brain read, and where does it write?
The Brain reads your meetings in Zoom and your documents in Notion and Google Drive, through connections you make and can remove. It writes ordinary pages into your own Notion or Drive, with each value's source passage one click away, and the pages stay yours if you leave. A workspace lives in the United States or the European Union, chosen once when it is made.
Who approves what goes into the record?
Your own people. The Brain shows them only what it is unsure of, they check it against the words it came from, and they approve or correct it in one screen. Every value keeps its history: what it replaced, when, why, and on whose word.
The readiness check
Where to start, and who sees what.
What is the Agentic Readiness Check?
The Agentic Readiness Check is Kelright's short assessment of how ready an engineering team is to work with coding agents. It takes about six minutes, asks a few lines about the company and then fourteen questions on how the team ships, tests and reviews and how it uses coding agents today, and needs no account.
What the check asks and gives backWhat do I get back from the readiness check?
A written read of which phase the team is in and the three places to look first, on the page as soon as the answers are sent and by email, short enough to forward. It is a read, not a score.
Who sees my readiness check answers?
When the check is taken from Kelright's own link, Kelright sees the answers and uses them to write the read, nobody else sees them, and Kelright erases them if asked. If a fund sent the link, the check says so before the first question and says who sees what: the fund sees combined numbers across the companies that answered, and a company's own answers only if the company allows it.
Still asking
Find out where your team stands, or ask us directly.
Prefer a link by email?hello@kelright.com
Opens an email to hello@kelright.com. Ask what you would have asked on a call.