Built to outlast whoever built it
Should we build AI in-house or bring in an outside firm?
Use your own people for what they know best and a specialist firm for the build and its upkeep. An outside firm can recommend changes an internal team can't easily make, and it stays accountable for what it delivered after the people who championed the project have moved on.
Owners
- Kyver
- Primary and backup engineer, shared on-call
- Internal
- AP manager owns the process; IT lead owns access
Documents, in your repository
- Architecture notesdocs/architecture.md current
- Runbookdocs/runbook.md current
- Agent policypolicy/ap-agent.yaml current
- Transition plandocs/transition.md current
- Model deployed
- extract-v3.2, hosted
- Last model update
- 3 weeks ago, after the regression suite passed
- Next review
- In 9 weeks, quarterly with the AP manager
How Kyver handles AI maintainability
An outside view
We have no stake in which department owns what. The assessment recommends what the numbers support.
Documented by default
Every system ships with architecture notes, runbooks and its policy files, kept in your repositories and accounts.
Managed and improved
We run, monitor and update what we build. Models and tools change every few months, and keeping yours current is our job.
Your team stays in the room
Internal staff take part in the build and own the decisions. We take on the upkeep, so their time goes back to the business.
Your star engineer will move up
Plenty of companies have a technophile or two on staff who could build this, so paying a firm looks unnecessary. There are two problems with that.
First, an internal team's motivations aren't necessarily yours. It works inside departmental politics, sacred cows and traditions that may no longer fit the business, and its recommendations avoid them. Second, brilliant people get promoted or leave. Whatever they built becomes unsupportable unless a replacement star happens to arrive at the right time.
Built in alongside maintainability
- SecurityCatch AI attempts to reach data or take actions outside their permissions, and block them before they run.
- GovernanceDecide which AI actions need a human sign-off, and make sure nothing irreversible happens without one.
- CostRun most AI work on hosted models that aren't metered by the token, so the monthly cost is predictable.
- TraceabilityRecord every AI session step by step, so failures can be investigated and fixed instead of guessed at.
Questions about AI maintainability
Can our internal IT team maintain what Kyver builds?
Yes. Everything is documented and lives in your accounts. Most clients keep us on to manage and update it, because models and tooling change constantly.
What happens if we stop working with Kyver?
You keep everything: code, configuration, policies, traces and documentation. We hand over operations with a written transition plan.
Why not hire an AI engineer instead?
For some companies that is the right call. One hire is a single point of failure, though, and experienced AI engineers are expensive. A firm gives you a team, coverage when someone is out, and continuity when people change roles.
Start with the assessment
Fixed price, agreed up front. Implement the plan with us within 6 months and we refund the fee.