AI agents hit real-world limits this week
AI agents moved from demo to policy fight this week. Several stories show where companies will welcome them and where they will push back.
This roundup covers five separate stories. Each one points to a different kind of control.
- United blocks outside AI booking agents.
- ChatGPT adds verified Business Agents.
- Adobe opens its apps to third-party agents.
- Google DeepMind releases AlphaGenome 2.
- Nvidia’s CEO resigns.
United says no to personal booking agents
United told reporters it will not let personal AI assistants book flights. According to The Verge, the airline cites privacy and regulatory worries.
The airline also says few customers have tried agent booking so far. Other businesses, such as Doordash, report similar traffic.
Why the stance matters
Airlines hold sensitive traveler data. They also face strict rules on refunds and changes.
An unverified agent makes accountability murky. If it books the wrong fare, who owns the mistake?
Still, the policy may also protect United’s own sales channels. Both motives can be true at once.
ChatGPT adds verified Business Agents
OpenAI took the opposite approach. The Verge reports that ChatGPT will let you chat with verified agents that represent a brand.
Early use cases include comparison shopping. The beta starts with eligible e-commerce and online service firms in the US.
Verification is the product
The word “verified” does the heavy lifting here. A brand-run agent answers for the company, so users know who stands behind the advice.
That contrasts with United’s worry about anonymous agents. Brands appear to prefer agents they can vet and control.
Adobe goes all in on agentic design
Adobe used its Max conference in Miami to push agents hard. As The Verge describes, your preferred third-party agents can now access and change work inside Adobe tools.
Adobe also showed a Firefly assistant and support for MCP. The goal is to automate repetitive creative tasks.
What creators should test
Automation helps most with batch work. Think resizing, versioning, and file cleanup.
Before trusting it, check three things:
- Can you undo every agent edit?
- Does the agent keep layers and history intact?
- Does your license cover the generated assets?
Results must repeat before a studio adopts any of this. A clever one-off demo is not a workflow.
Creators who edit on the go may also want a good color-accurate monitor (paid link) for reviewing agent output in detail.
AlphaGenome 2 studies disease at cell level
Google DeepMind unveiled AlphaGenome 2. The Verge says the model predicts how DNA changes affect different cell types.
That detail matters. The same variant can act differently in different cells.
Better predictions could point researchers toward causes of disease. They could also help guide new treatments.
This is science, not consumer software. Lab testing will decide how useful the model proves.
Nvidia’s CEO resigns after scandal reports
Nvidia’s board separated from co-founder Jensen Huang, according to The Verge. Reports described two extramarital affairs, including one with an alleged Chinese spy.
The company announced his exit as CEO and president after a late board meeting. Investors will now watch for a successor and for any change in strategy.
Nvidia supplies the chips behind most AI work. Leadership turmoil there reaches every tool in this roundup.
Takeaways by user type
The common thread is trust. Each company decides which agents it will let in.
- Shoppers: expect verified brand agents before open-ended ones.
- Creators: test Adobe’s agent access on low-risk projects first.
- Investors: watch Nvidia’s succession plan closely.
- Science watchers: follow independent studies of AlphaGenome 2.
Agents will spread where identity and accountability are clear. Where they are not, doors will stay shut.
As an Amazon Associate, TechMogo earns from qualifying purchases.
