Science & Tech · AI · 8 hrs ago
Businesses Urged to Set Limits Before Deploying Autonomous AI Agents
Autonomous AI agents can carry out business tasks without waiting for instructions, including sending emails and processing invoices.
The advice came from a discussion involving AI, finance and technology leaders, including representatives of Box, Deliverance AI, Ralio and Geordie AI.
Unlike copilots that respond to prompts, agents can make decisions and act on their own, sometimes using company databases and systems.
That speed raises the stakes: one mistake could quickly affect business operations, especially if an agent has wide access.
Speakers said companies should first organise their data and give agents clear instructions, limited permissions and ongoing human oversight.
Businesses should not give agents full autonomy from day one, and governance, compliance and risk teams should be involved before pilot projects begin.
Autonomous AI agents can access databases, carry out tasks and make decisions without human prompting, raising the risk that a mistake could quickly affect business operations.
Business leaders should set strict limits, organise the data agents use and maintain oversight before deploying them, panellists said at a Sifted Talks discussion.
The panellists compared granting agents autonomy to onboarding employees, with trust and access increasing over time rather than being given immediately.
They also said humans remain accountable for what agents do, and that governance, compliance and risk teams should be involved before pilot projects begin.
- Who
- The advice came from panellists including Mick McNeil of Deliverance AI, Eléonore Hébrard of Box, Leo Rosales of Ralio and Hanah-Marie Darley of Geordie AI.
- What
- Businesses were urged to set boundaries and oversight before deploying autonomous AI agents.
- When
- The discussion was described as Sifted’s most recent Talks; the article was published on 2026-11-09.
- Where
- Not stated.
- Why
- Agents can act quickly and access business systems, so errors may cause significant harm; disorganised or outdated data can also lead to failures.
This story does not have two clearly opposing sides.
This story does not have a timeline yet.
- Discussion
- Sifted Talks, in partnership with Box
- Mick McNeil
- Founder and CEO of Deliverance AI
- Eléonore Hébrard
- Solutions engineer at Box
- Leo Rosales
- Cofounder and CTO of Ralio
- Hanah-Marie Darley
- Cofounder and chief AI officer of Geordie AI
1 source
What people are saying
At no point on day one with AI agents should you give full autonomy across business management systems.
You don't have an AI problem, you have a content problem.
We [at Ralio] follow zero trust principles. We assume the model is compromised and we build around it.