Context and planning
- Every repo has a context file with what the team needs, not one person’s setup.
- Auth, infrastructure and data-model changes are always planned first.
AI Native Delivery
Built with AI.Owned by people.
Our developers work with AI coding agents every day. One shared standard decides how: plan first, run agents safely, and a human reads every line before it reaches you.
The honest part
AI writes code fast. That was never the hard part.
Unsupervised AI produces plausible code, tests that test nothing, and diffs nobody can review. The speed is real. So is the mess, unless someone owns the output. At Xelerit the rule is simple: AI-generated code is our code. “The model wrote it” never explains a bug.
What you get
The same team, going further.
Speed
Faster first versions
Features, refactors and migrations start from an agent draft, not a blank file.
Quality
More testing for the same budget
Agents scaffold the tests; our engineers check what they actually assert.
Knowledge
Documentation that stays current
Context files, specs and decision records are updated as part of the work.
Continuity
Small, stable teams
The people who know your product stay with you. AI makes each of them go further.
How a feature moves from idea to merge
Spec first. Then code.
Every change is written down before it’s built, and every step has a person who signs it off.
Spec
We write down what we build and why. A developer approves it, never the agent. Major features get their spec reviewed on its own.
Plan
The spec becomes tasks, each with the tests that prove it works.
Check
The plan is cross-checked against the spec and the project’s conventions. Open questions come up now, not halfway through.
Build
Agents work task by task in isolated workspaces, test-first, with a review after each task.
Deliver
One clean, reviewable pull request with the docs included. A developer opens it. The agent never does.
Close
The spec is marked as implemented, with a date. The decision stays on record for the next person.
We adapt this flow to each project and don’t impose a framework. What never changes: every change gets written down.
Our standard
One shared standard, written from what already works.
How to read it: we took what works across our projects and wrote it down in one place. MUST means binding. SHOULD means a strong default. The project’s own conventions always win.
Running agents safely
- No blanket auto-approve, ever. Destructive commands are denied by default.
- No production credentials in an agent environment.
- Agents run in a sandbox or container.
Commits and pull requests
- Protected branches, pull requests, green CI.
- No unreviewed agent commits.
- Diffs a human can actually read, so big changes get split.
Code review
- You read your own diff before anyone else does.
- AI is a first pass, never a replacement for a human reviewer.
Testing and data
- We read what the tests actually assert.
- An agent never goes green by weakening a test.
- Customer data never goes into fixtures or an agent’s context.
Tooling and security
- Only the skills and integrations a task needs.
- Supply-chain protection on every package install.
- OWASP Top 10, plus OWASP Top 10 for LLM Applications when we build AI features.
- An extra review pass on anything touching auth, input handling or secrets.
In your environment
Your environment, your rules. Our recommendation, always on the table.
Where the environment is ours, our standard is binding. Where it’s yours, we recommend it, make the case, and come back to it. Saying no is your call. Never raising it would be our failure. Every exception is documented and approved under our Secure Development Policy.
Security and ownership
- ISO 27001 certified. Our AI guidelines sit next to our information security policies.
- Your code stays yours. It never trains a public model.
- Full ownership. Code, specs, prompts and documentation belong to you.
Who it’s for
- ISVs and product companies whose roadmap is bigger than their team.
- Teams already using AI that want it done safely and consistently.
- Companies extending their team with Xelerit developers who bring an AI-native practice with them.
FAQ
Questions we hear about AI in delivery.
- We are. The developer who opens the pull request has read every line, and a human reviews it before it merges.
- No. We agree how AI is used with you first. In your environment your rules apply, and every exception is documented and approved under our Secure Development Policy.
- The right model for each task, chosen within our shared standard and your environment’s rules. We test new models and tools as they appear and keep what works.
- Yes. It’s written down to be shared, and we can walk your team through it as part of working together.