What is AI stewardship? The work after launch.
AI stewardship is the ongoing work of monitoring, maintaining and improving AI systems after launch, so they stay accurate, stay current as tools change and keep paying back. Modern Thesis, an AI strategy and implementation firm for non-tech companies, provides it with no set end date and reports results against the baseline measured before the build.
What does AI stewardship include?
AI stewardship covers everything that keeps an AI system accurate, working and worth its cost once it is in daily use. Building the system is a project. Looking after it is a responsibility, and it falls into five parts.
- Monitoring. Someone reviews a sample of real output on a schedule, tracks error rates and looks at every case the system flagged for a person to decide.
- Maintenance. When an AI provider retires a model, a connected system changes its software or a login expires, someone fixes it before the team notices.
- Improvement. As better models and features arrive, someone tests them against the current setup and switches when the results are better.
- Adoption. New hires get trained, and the steps people skip or work around get fixed.
- Reporting. Leadership sees the return in the same numbers measured before the build.
Why do AI systems need ongoing care after launch?
AI systems age faster than most business software, because several things they depend on keep moving. A system that worked well in March can give worse answers by September without anyone touching it. There are four common causes.
The models change
AI providers retire older models on a published schedule. Anthropic's documentation says it gives at least 60 days' notice before retiring a publicly released model, and that requests to a retired model fail. Claude Sonnet 4, launched in May 2025, was retired on June 15, 2026.1 A replacement model is usually better overall, but it can behave differently on your specific task, so someone has to test it on your work before the switch.
The tools improve
A step that needed custom engineering last year may now be handled better by a newer model, or by a feature inside software you already pay for. Without someone watching, a business keeps paying to maintain the old way.
The business changes
New services, new pricing, a reorganized team, a new form from a regulator or a key employee leaving can all change the workflow the system was built for. The system keeps doing what it was told, even after that stops being right.
Accuracy needs watching
AI output varies in ways ordinary software does not, and small slips add up quietly. MIT's 2025 study of AI in business found that "most GenAI systems do not retain feedback, adapt to context, or improve over time," and buyers in its interviews ranked the ability to improve over time among the things they look for in a vendor.2 Stewardship is how a system gets that improvement: people look at the mistakes and change the system so they stop.
How is ROI reported during stewardship?
ROI is reported against the baseline measured before anything was built: how long the workflow took, what it cost and how often it went wrong. The same measures are taken again after launch, the same way, so the comparison is fair. Results go into numbers the business already tracks, such as hours per job, quotes sent per week, days to close the month, rework or overtime. A useful report fits on one page. It shows the baseline, the target, the current figure, what the system cost to run in the period and what changed. If a number goes the wrong way, the report says so.
How is AI stewardship different from IT support or a software maintenance contract?
IT support keeps your technology running, a maintenance contract keeps a vendor's product working, and AI stewardship keeps the results of systems built for your workflows accurate and paying back. Most companies at this size need IT support either way. Stewardship sits alongside it.
| IT support | Software maintenance contract | AI stewardship | |
|---|---|---|---|
| What it looks after | Devices, networks, accounts, security | One vendor's product | AI systems built around your workflows |
| What sets it in motion | A ticket or an outage | A bug report or a scheduled release | A schedule of reviews, plus changes in models, tools or the business |
| How success is measured | Uptime and response time | The product works as specified | Accuracy and return against the baseline |
| Who decides what changes | Your IT provider | The vendor, for all its customers | You and the steward, for your business |
The difference that matters most is the last row. A software vendor improves its product for the average customer. A steward improves your system for how your company actually works.
Why doesn't AI stewardship have an end date?
Stewardship has no set end date because the reasons for it don't end. Providers will keep retiring models, better tools will keep arriving and the business will keep changing. A fixed term suggests the system will one day be finished, and AI systems in daily use are never finished in that sense. Each review also tends to surface the next opportunity, so stewardship feeds the next round of strategy.
What should you ask any provider about post-launch support?
- Who checks the accuracy of the output after launch, and how often?
- How will you know when an AI model we depend on is being retired, and who tests the replacement?
- What happens when one of our connected systems changes its software?
- How will results be reported, and against what baseline?
- Whose name are the accounts in, and who owns the code and documentation?
- What is included each month, and what costs extra?
How Modern Thesis handles this
Stewardship is one of the four disciplines Modern Thesis covers, alongside strategy, people and implementation. The same team that builds a system looks after it. We monitor and maintain what we build, keep it current as better tools arrive and report the return against the baseline from the strategic assessment. It has no end date. The client owns its data, accounts, code and documentation throughout.
Questions
What is AI stewardship in simple terms?
AI stewardship means someone stays responsible for an AI system after it goes live. They check that its output is still accurate, fix it when a connected system or AI model changes, improve it when better tools arrive, and report what it returns to the business. The word comes from the idea of looking after something on the owner's behalf. The owner keeps the system and the data. The steward keeps it working and keeps it worth having.
How often should an AI system be reviewed?
Check accuracy continuously on a sample of real output, and review the business results with leadership on a fixed schedule. A sensible pattern is close attention in the first months after launch, when most surprises appear, then a monthly or quarterly review once the system settles. Review it again whenever something upstream changes: a new AI model, a software update in a connected system, a new product line or a change in how a team works.
Do we need stewardship for off-the-shelf tools like ChatGPT or Copilot?
The vendors maintain the tools themselves, so you don't need anyone to keep ChatGPT or Copilot running. What they don't maintain is anything you built around them: custom instructions, connections to your systems, automated workflows and the agreed way your team uses them. Stewardship matters most for systems built for your business, because nobody outside your company is watching whether those still give the right answer.
If stewardship has no end date, are we locked in?
No end date describes the work, which keeps going because models, tools and your business keep changing. It shouldn't mean dependence. At Modern Thesis the client owns its data, accounts, code and documentation, so the systems belong to the business and can be run by whoever the business chooses. Ask any provider the same question before you sign, and get the answer in writing.
What does AI stewardship cost?
It depends on how many systems are running, how many tools and data sources they connect to, and how much improvement work the business wants each period. Modern Thesis prices each engagement to its scope after a first conversation, and stewardship is planned with the build, so its cost is known before launch. Whatever you pay, judge it against the return reported from your baseline.