Employees spend 6.4 hours a week 'taking care' of AI — and 75% claim that productivity increases

Workers are dedicating, on average, 6.4 hours a week to 'care' artificial intelligence tools — tasks such as providing context, debugging results and correcting errors. This is what points to a research published by the GovTech portal, which also shows that 75% of professionals believe that AI increases their productivity, even with the cost of time.

What's botsitting? Understand the cost of time and how to reduce

The term botsitting (a combination of bot and babysitting) describes the work of monitoring and adjusting AI models to deliver useful results. In practice, this means writing more accurate prompts, checking the quality of responses, and correcting hallucinations — errors in which AI invents information. GovTech's research shows that this 'digital babe' consumes almost a day's work per week, at a time when companies run to integrate generic AI into their operations.

How to reduce botsitting in your company? Practical tips

The reduction of botsitting time goes through two fronts: better data preparation and employee training. Tools such as the Anthropic Claude Fable 5 model, which balances power and security (see details on Anthropic reveals Claude Fable 5: the Mythos model that balances power and security for general use), help minimize manual corrections. In addition, more robust infrastructure — such as those tested in the NVIDIA Blackwell-led benchmark (NVIDIA Blackwell leads first infrastructure benchmark for AI Agents) — can reduce the need for human intervention. The search also recommends creating tested prompts repositories and document common errors.

Botsitting versus productivity: does AI justify the cost?

Yes, according to the employees themselves. Despite the 6.4 hours per week, 75% of the interviewees stated that AI makes them more productive. This suggests that the time spent on botsitting is perceived as an investment, not a waste. However, experts warn that if maintenance time grows uncontrolled, the return on investment may fall. Recent cases such as the revolt of developers with excessive Anthropic guardrails (Anthropic faces revolt with Fable 5: excessive guardrails choke developers show that the balance between autonomy and control is delicate.

Botsitting and productivity: what research reveals

GovTech's research heard professionals from various sectors and found that botsitting is not an isolated phenomenon. The most common tasks include: writing and adjusting prompts (36% of the time), checking answers (28%) and correcting factual errors (22%). Only 14% of the time is dedicated to training new models. This indicates that, in the short term, most of the effort is to ensure that existing AIs work properly, not to expand their use.

Frequently Asked Questions about botsitting and AI maintenance

How many hours do employees spend on botsitting?

On average, 6.4 hours a week, the equivalent of almost a full day’s work — considering a 40 - hour day work.

All this time is wasted?

No, 75% of respondents claim that AI makes them more productive, suggesting that time spent is an investment necessary to obtain quality results.

How to reduce botsitting time?

Improving data curation, using well-adjusted templates, creating validated prompt libraries and investing in team training are effective strategies pointed out by the research.

Fonte: GovTech - Question of the Day