Sunday, November 22, 2026

Top 5 This Week

Related Posts

AI agents: seven IT managers call for their actions to be regulated | Brief AI


Seven technology executives surveyed on 2027 priorities recommend limiting the actions of AI agents and retesting when their model changes. Their recommendations also cover measuring results, data preparation and choosing between external and self-hosted models.

Define the responsibilities and rights of agents

At insurance claims management specialist Sedgwick, Sean Safieh leads information systems for global platforms and digital solutions. He describes agents who are no longer limited to summarizing or recommending: they guide the work and trigger transactions. A sales team can, for example, allow an agent to validate a transaction alone. For this manager, these capabilities present a different operational risk, particularly in complex organizations.

Sean Safieh contrasts the generally identifiable responsibility of a person with that of an agent, for whom it is necessary to determine who authorized access and what acts are permitted. It recommends explicit monitoring, auditable and explainable actions, as well as reliable traceability. At Sedgwick, he explains, AI governance is a matter of business, not just IT control: it touches on risks, compliance and trust in tools. Before increasing the autonomy of an agent, he recommends treating it as an identity of the company and defining responsibilities with the commercial and risk departments.

For Boris Kolev, global head of technology at JA Worldwide, an NGO dedicated to youth education, change is also about the actions allowed, and not just the reliability of the answers. According to him, modifying data, initiating a transaction, communicating with the outside world or activating another system transforms an error into an operational risk. It requires management of identities, permissions and audit trails adapted to agents. Expenditure ceilings, escalation procedures to a manager and the possibility of human control should, in his eyes, precede their deep integration into critical processes.

Retest systems when their model changes

Frank Meltke, CEO of digital transformation consultancy Contraco, warns that an already approved system can change models without new authorization. Vendors are continually replacing and updating their models, he says. An assistant validated in the spring could therefore behave differently in the fall while retaining his name, his rights and his role in the processes. He believes that very few companies then trigger a new review.

The evening letterFree · 7 p.m.

Do you want the best AI tools before the rest?

We test and decipher the new AI tools every evening, in 5 minutes. Free.

Included upon registration: our selection of the best AI guides & comparisons.

Choose your pace

Free · No spam · Unsubscribe in 1 click

His recommendation is to register the model and its version for each system in production, then ask vendors to announce changes in advance. Initial acceptance tests should be replayed with each change, he insists, especially when a tool can approve an operation, make a purchase or modify records without human intervention.

Daniel Herrington, director of AI at insurance comparator The Zebra, calls for the development of synthetic evaluations to check how agents act on the data. He considers companies that have not invested in it to be already behind the times. He considers it naive to rely on traditional protections such as two-factor authentication or CAPTCHAs and predicts that security platforms will eventually lose this face-off.

Measure results and prepare data

AI is among the top concerns of information systems managers for 2027. For Michael Corrigan, who holds this position at World Insurance Associates, an investment must respond to a specific problem, commercial or IT. Before deploying the tool, the company should define what it will measure and the criteria that will be used to judge its effectiveness. The same indicators must then make it possible to compare the previous situation with the new results, he recommends.

Mojgan Lefebvre, director of technology and operations at insurer Travelers, emphasizes data preparation. She reports that an experienced specialist on her team considers their good organization as a prerequisite for using AI. Quality, classification and understanding must be worked on, she explains. It also highlights the need to restore the commercial context of the data, as their interpretation by AI differs from that of humans.

Distribute requests between models

Andy Boettcher, chief innovation officer at DoubleTrack, an IT services and consulting company, anticipates a mix of large, cutting-edge models and open models hosted by companies themselves in 2027. In this scenario, routers would direct each request according to classification rules. It presents the management of this whole, rather than a single model, as a challenge to prepare for.

The manager says he uses three open models on his machine alongside a cutting-edge model and judges their performance to be close to the latter. He recommends defining now which data and which questions can be entrusted to each destination. The objective he sets is to keep the most sensitive work within the company.

Shredder Smith
Shredder Smith
Shredder Smith is the lead curator and digital persona behind topaitools4you.com, an AI directory dedicated to "shredding" through industry hype to identify high-utility software for everyday users. Smith positions himself as a blunt, no-nonsense reviewer who vets thousands of emerging applications to filter out overpriced "wrappers" in favor of tools that offer genuine ROI and practical productivity. The site serves as a watchdog for the AI gold rush, providing categorized rankings and transparent reviews designed to help small businesses and creators navigate the crowded tech landscape without wasting money on low-value tools.

LEAVE A REPLY

Please enter your comment!
Please enter your name here

Popular Articles