Who runs the AI in your company: why a business needs a Claude Code operator
Companies pay for AI subscriptions and commission rollouts, and the effect turns out smaller than promised. The reason is rarely the model: what is missing is a person who sets the AI its tasks every day and answers for the result.
Most companies already have artificial intelligence. Someone paid for staff subscriptions, someone commissioned a chatbot or a voice assistant from an integrator, someone connected AI to email and documents. A few months later the manager admits it honestly: the effect is smaller than promised. The bot quotes outdated prices, the assistant "does something", but nobody can say how many hours it saved, and half the staff have gone back to their old spreadsheets.
The reason is rarely the model itself. Almost always what is missing is a person who sets the AI its tasks every day, accepts the result and answers for it the way a chief accountant answers for the books. We call that role the operator and in this article we explain, using Claude Code as the example, why it is needed even after the best possible rollout.
What the figures say
of working hours saved by AI on average across all US employees - including those who do not use it
St. Louis Fed, Q3 2026more enquiries per hour where AI was built into the support desk
NBER, "Generative AI at Work"The broadest picture comes from the quarterly tracker of the Federal Reserve Bank of St. Louis. By its figures, in the third quarter of 2026 generative AI saved US workers on average about 2.1 % of all working hours. That is an average across everyone employed aged 18-64: those who do not use AI count as zero, which is why the overall figure looks modest.
The result is quite different where AI has been built into a specific job. The study "Generative AI at Work" (Brynjolfsson, Li, Raymond, NBER) followed the rollout of an AI assistant to 5,179 support agents: enquiries resolved per hour rose on average by 14 %, and among newcomers and the least experienced by 34 %. These are different measurements, so one cannot be divided by the other. But the direction is the same: the effect appears where AI is built into a specific job.
And the MIT NANDA report "The GenAI Divide" (2025) explains why not everyone manages to build it in: by its count, only about 5 % of corporate AI pilots produce a measurable financial result. The authors see the cause not in the quality of the models but in the fact that the tools are not embedded in the company's working processes and do not adapt to them. The problem, in other words, is managerial rather than technological.
The effect comes not from access to AI but from the person who turns it into a process.
Claude Code is not a chat window but an agent inside your working environment
Most people know AI as a chat window: question, answer. Claude Code is built differently. It is an agentthat works inside the company's own environment: it reads the project and the documents, proposes a plan, makes the changes itself and shows exactly what it changed. Through connectors - ready-made links - it is joined to email, the calendar, accounting systems and the website, and then it works not with a single file but with the real data of the business.
accepts the result and answers for it
That is exactly why it is more powerful than a chat window - and exactly why it needs directing. An agent that can change a page on the site, a price in the database or the text of a contract must have an owner: someone who decides what it is allowed to do, and who checks the work before it reaches a client.
Which tasks can already be handed to an agent inside a company
We work with Claude Code every day on our own products - Smart CMS, Smart 360, Smart LMS, BOT-panel - and on client websites. These are the tasks the agent already performs reliably for us, under an operator's direction:
Websites and content
Move a site off an old system, check every page against the original on desktop and phone, translate a site into several languages, update the news section.
Documents
Draw up a contract or an acceptance act from the company template, reconcile payments against a bank statement, prepare a register of documents.
Email and calendar
Work through the inbox, prepare draft replies, find what was agreed in a thread. A person sends them.
Chatbots and voice assistants
Update the bot's knowledge base when new products appear or prices change, and test the answers with control questions.
Infrastructure
Check a server for vulnerabilities, set up backups, watch for failures in automation scenarios.
Reports
Pull data from several systems into a single report for the manager.
The operator does not need to be a programmer - they need to know their own area and to be able to set a task so that the result can be accepted.
Where an agent fails quietly
The most dangerous AI mistakes are not the ones where it refuses to work, but the ones where it confidently reports "done". A few situations from our own practice:
"Done" is not the same as "finished"
The agent completed the page layout and showed green checks. On a phone the block ran off the edge of the screen.
Operator's ruleOpen the page on a phone before accepting it. Automated checks do not see this.
Literal execution
Tell the agent "replace the phone number everywhere" and it will do so flawlessly - even if the new number in the instruction is wrong.
Operator's ruleState the old and the new value as digits first, and only then start a mass change.
A mass change with no way back
A single inaccuracy in a replacement pattern can spoil dozens of database records in a second.
Operator's ruleBefore any mass action, a backup. Then a mistake costs minutes rather than a day.
Knowledge that has gone stale
For several weeks a voice assistant was quoting customers the old tariff names: after the renaming, nobody updated its instructions.
Operator's ruleIf prices or names change, the agent's instructions are updated the same day.
In every case what saves you is not a "cleverer model" but a person with a clear working cycle.
Why an operator is needed after the integrator
An integrator runs a project: they install the system, configure it, train people and hand it over. But the system goes on living. Prices change, products appear, new people arrive, data and security requirements change. Without an owner, even a perfectly implemented AI starts falling behind the company's reality within a few months.
| Integrator | In-house operator | |
|---|---|---|
| Format | A project with an end date | Daily work |
| Answers for | Launching the system | The system's result every day |
| Knows | The technology | The technology and their own department's processes |
| When prices or the product change | A new order | Updates the same day |
| Access and client data | Hands over after launch | Holds and controls it |
This is not a rivalry. A good integrator positively wants the client to have someone the system can be handed to - otherwise in six months they are called back to "fix the AI that broke".
The operator's cycle: four steps on every task
Task
"Make it look nice" becomes the boundaries of the change and a test of completion: what exactly we are changing, what we leave alone, how we will know it is done.
Plan
The agent shows the steps before it changes the first line. A wrong direction is stopped here, not afterwards.
Diff - the list of changes
The operator reads the changes themselves, line by line, not the agent's summary.
Acceptance
"Done" is said after the checks - on the live page, on a phone, on real data - not instead of them.
Until this cycle becomes a habit, the agent stays a toy. Once it has, it is a tool that takes hours of routine off the team every week.
From SmartGT Academy
How we teach this
We teach this role at SmartGT Academy. The course is not about the tool but about directing it: six weeks of practice on the participant's own tasks.
Claude Code Operator
Programming is not required. For SmartGT clients - UAH 15 000; a corporate group of up to 8 people - UAH 120 000.
The cohort is led by Mykhailo Vasianovych, founder of SmartGT, who works with Claude Code daily on the very products that become the practice material.
Ranks: a qualification that grows with practice
A certificate saying "attended the course" says nothing about what the person will do on Monday. So for operators we keep a scale of ranks: ten steps, each confirmed by hours worked with the agent, by the number of working sessions, by the tokens spent and by a practical task.
All ten ranks and their thresholds
| Rank | Threshold | What the person does unsupervised |
|---|---|---|
| R0 · Connected | first dialogue | The tool is installed and the permissions are set |
| R1 · Trainee | 6 h · 3 sessions | Tasks under supervision; stops the agent on a wrong direction |
| R2 · Operator | 14 h · 7 sessions | Runs a task alone: from the brief to an accepted diff |
| R3 · Reviewer | 26 h · 13 sessions | Checks the agent's work done for others |
| R4 · Skills master | 42 h · 21 sessions | Fixes the department's routine into the team's tools |
| R5 · Integrator | 60 h · 30 sessions | Connects the agent to email, the calendar and accounting systems |
| R6 · Sentinel | 80 h · 40 sessions | Answers for access, secrets and the isolation of client data |
| R7 · Conductor | 95 h · 48 sessions | Runs several agents in parallel and can train operators |
| R8 · Architect | 180 h · 90 sessions | Owns the process in the department: branches, checks, procedure |
| R9 · Mentor | 350 h · 175 sessions | Trains operators and holds the standard across the company |
Participants reach the first two or three ranks during the course itself. After that the rank grows with the work. Each rank has a badge with a permanent code and a page of its own that an employer can open and check. For a company that is a clear criterion: a vacancy can name the rank required.
Mykhailo Vasianovych · R8 "Architect"the cohort trainer's badge
Open the badge →A rank is not bought; it is earned.
Where to start
Choose one task
The one that takes someone on the team several hours every week.
Choose a person
Someone who knows that area well. This is your future operator.
Close that task on the course
Together with us; the operator will run the next ones alone.
Train an operator, or a whole team at once
Terms used in this article
- Agent
- an AI-based program that does not merely answer but carries out actions itself: reads files, makes changes, runs checks.
- Operator
- the person who sets the agent its tasks, checks the changes and answers for the result.
- Connector
- a link joining the agent to a working system: email, calendar, accounts, the website.
- Diff
- the list of what the agent actually changed: before and after.
- Token
- a unit of the volume of text a model has processed; in the ranks, a measure of accumulated work.
Sources
- Federal Reserve Bank of St. Louis, FRED: Generative Artificial Intelligence, Time Savings: Employed Adults (RPSGENAITSALL), data for the third quarter of 2026.
- Erik Brynjolfsson, Danielle Li, Lindsey R. Raymond. Generative AI at Work, NBER Working Paper 31161.
- MIT NANDA, "The GenAI Divide: State of AI in Business 2025", as reported by Virtualization Review.
SmartGT is an IT team providing complete business solutions. SmartGT Academy is an independent training project. Claude and Claude Code are trademarks of Anthropic; we are not an official partner of the company.