Cloud Combinator Daylight Factory

Software built in daylight

Agentic engineering, measured on your own delivery data.

Daylight Factory is a governed agentic engineering platform that runs in your own AWS account. Agents take scoped tickets, write and test the change, and open a pull request. Nothing ships until a named person signs it off. In 8 weeks you know what it’s worth, measured on your own delivery data.

Book a demoSee how it works
  • AWS Advanced Tier Partner
  • Runs in your AWS account
  • A person approves every merge
  • A go or no go decision in 8 weeks

Your engineers got faster. Your delivery numbers didn’t.

Most engineering teams now have an AI assistant in the editor. Few can show the difference in what ships. The work moved, it didn’t shrink.

  • Code is written faster, then waits in review.
  • Nobody can say what a change made by an agent costs, or whether it was worth it.
  • Security can’t see what the agents were allowed to touch.
  • Every engineer runs a different tool with a different setup, so nothing repeats.

Daylight Factory fixes the system around the agent: the tickets going in, the checks on the way through, and the approval at the end.

Four stations. The last one is always a person.

Daylight Factory isn’t “let the AI write everything”. It’s a pipeline where agents handle scoped changes and people approve each one.

  1. Scope. A ticket says what changes, what done looks like and what’s off limits. We work with your team until tickets are clear enough for an agent to act on.

  2. Build. An agent picks up the ticket, writes the change, writes the tests and documents what it did. It runs as an isolated workload inside your AWS account.

  3. Check. Your automated tests, security scans and policy gates run on every change before a person sees it.

  4. Approve. A named engineer reviews the pull request and approves the merge. Branch protection enforces it, so no agent merges its own work.

Every change leaves a trail in your own Git history and audit logs: what the agent did, which checks ran, and who approved it.

What’s inside the platform

Four components, all inside your AWS account
Four components, all inside your AWS accountEverything sits inside your AWS account. The agent runner, where agents work scoped tickets in isolated runs, sends its model calls to the model gateway, which applies budgets, rate limits and audit to every model call. Each change goes from the agent runner to the policy gates, which run tests, security scans and policy checks, and then to human approval, where a named engineer approves every merge. Results from the policy gates and from human approval go to the scorecard, which tracks merge rate, cycle time, review time and cost per merged pull request every week.Your AWS accountModel gatewayBudgets, rate limits andaudit for every model callAgent runnerAgents work scoped ticketsin isolated runsPolicy gatesTests, security scans andpolicy checks per changeHuman approvalA named engineer approvesevery mergeScorecardMerge rate, cycle time, review timeand cost per merged pull request,every weekmodel callsresults

Agents run in the agent runner and reach models only through the gateway. Every change passes the policy gates and a named engineer before it merges, and the scorecard records the result.

What’s yours by week 8

  • A platform in your AWS account. Your code stays in your repositories and your AWS account. Model access goes through one governed gateway on Amazon Bedrock, with budgets, rate limits and audit logs per team and per user.

  • Your engineers’ own tools. Claude Code, Cursor or any client that works with the OpenAI API connects through the same gateway. Nobody has to switch editor.

  • A delivery scorecard. Agent pull request merge rate, pull request cycle time, review time, cost per merged pull request and DORA metrics, tracked every week against your baseline. Change failure rate is the guardrail.

  • A maturity assessment. Where each team sits from L1 to L5 today, and what it takes to move up a level.

  • Engineers alongside yours. Cloud Combinator forward deployed engineers work inside your team for the whole pilot.

  • A decision. A go or no go report built on your numbers, not ours.

The parts nobody budgets for

Giving engineers an agent takes an afternoon. Running agents safely across a team takes all of this, and it’s in place from week one:

  • One gateway for every model call, with budgets and rate limits per team and per user
  • An audit log of who and what called which model
  • Isolated runs for every agent, inside your AWS account
  • Agents started straight from a ticket or pull request
  • Agent roles for product, architecture, development, review and operations
  • Branch protection that enforces human approval on every merge
  • Tests, security scans and policy gates wired into every change
  • A baseline of your delivery numbers before any agent runs
  • A weekly scorecard, including cost per merged pull request
  • A ticket standard your agents can act on

A platform, and the engineers to land it.

  • It runs in your AWS account.Your code, your data and your model usage stay under your own controls.

  • It comes with engineers.Our engineers deploy it inside your team.

  • It works with the tools you have.Your engineers keep their editors and agents. The factory governs and measures all of them.

  • AWS can fund part of it.As an AWS Advanced Tier partner we check your eligibility and handle the application.

Where is your team today?

We place each team on a five level scale. It’s our working model, not an industry standard. The pilot aims to take one team to L4.

  1. L1Assisted

    Engineers use autocomplete and chat in the editor.

  2. L2Paired

    An engineer and an agent work one task together.

  3. L3Supervised

    Agents complete whole tickets, with a person watching each step.

  4. L4Delegated

    Agents work the backlog on their own. A person approves each merge.

  5. L5Factory

    The pipeline runs across teams as the normal way of shipping.

Start with one team. Scale when the numbers say so.

Fixed fee for the 8 week pilot, then a platform subscription per team.

  • Pilot

    Onboarding for one team: the platform deployed and proven on real work in 8 weeks

    How it’s charged

    Fixed fee

    Pilot includes:

    • Readiness and maturity assessment
    • Platform deployed in your AWS account
    • Backlog scoping with your team lead
    • Forward deployed engineers for all 8 weeks
    • Weekly scorecard
    • Go or no go report

  • Platform subscription

    The platform for every team that goes live after the pilot

    How it’s charged

    Subscription per team

    Platform subscription includes:

    • The platform for each subscribed team, with new releases as they ship
    • Policy gates set for each repository
    • Monthly scorecard and maturity review
    • Training for team leads and reviewers

  • Managed

    The platform operated for you by Cloud Combinator

    How it’s charged

    Subscription per team

    Managed includes: everything in the platform subscription, with Cloud Combinator operating, monitoring and upgrading the platform for you.

AWS funding. Eligible customers can have part of the pilot funded by AWS. We check eligibility and handle the application.

AWS running costs. AWS bills infrastructure and model usage to your own account. We give you an estimate before the pilot starts, and team budgets cap the spend.

For software companies on AWS with more backlog than engineers

A good fit if:

  • You build software on AWS and your roadmap is bigger than your team.
  • Your engineers already use AI coding tools, and you can’t see the return.
  • You have automated tests and a CI pipeline.
  • You need to show who approved what, for customers, auditors or regulators.

Not there yet? If you don’t have automated tests or CI in place, start with Agentic Engineering. It takes a team through the first two stages, and Daylight Factory is where it leads.

Frequently asked questions

Where does our code go?

It stays in your repositories and your AWS account. Agents run inside your account and reach models through Amazon Bedrock.

Which models do the agents use?

Models on Amazon Bedrock, including Anthropic’s Claude. All access goes through one gateway, so you set budgets and rate limits per team and per user.

Can we start small?

Yes. The pilot is one team and one backlog. We begin with low risk work such as dependency updates, test coverage and small fixes, then widen the scope as the merge rate holds.

Do our engineers have to change tools?

No. Claude Code, Cursor and any client that works with the OpenAI API connect through the gateway.

How is this different from buying AI coding licences?

Licences make one engineer faster. Daylight Factory governs and measures the whole path from ticket to merge, whichever tools your engineers use.

What do we need before we start?

An AWS account, code on GitHub, automated tests in CI, and a team lead who owns the pilot.

What does it cost?

A fixed fee for the 8 week pilot, then a platform subscription per team. AWS bills infrastructure and model usage to your account, and we estimate both before you start.

What happens after week 8?

You get a go or no go report built on your own numbers. If it’s a go, the team moves onto the platform subscription and more teams follow.

See it on your own backlog

Book a 30 minute demo. We’ll show agents working a real backlog, from ticket to approved merge, then tell you whether a pilot makes sense and what AWS funding you could get.

No commitment. If you’re not ready for a pilot, we’ll say so.

We use cookies

This website uses cookies to enhance user experience and to analyze performance and traffic on our website.

See our Privacy Policy for details.