Silicon Valley GirlHead of ChatGPT & Codex: agents for normal people are HERE
At a glance
WHAT IT’S REALLY ABOUT
OpenAI’s Thibault previews reliable agents transforming everyday knowledge work
- AI agents are becoming reliable over long horizons and can use many tools (email, calendar, browser, apps), making benefits accessible to non-technical users soon.
- Knowledge workers (e.g., marketers) will increasingly automate repetitive research, inbox triage, and reporting via scheduled agent runs and “approve-this” dashboards.
- Safety and trust are addressed through mechanisms like “Auto Review,” where a second agent verifies a primary agent’s actions to reduce harmful or risky behavior.
- For personal data and context, examples beat explanations (e.g., provide writing samples instead of describing tone), and storage will shift from local files toward cloud-based memory and shared access.
- In software creation, “vibe coding” is great for prototypes, but scaling still benefits from technical oversight until agents improve maintainability over the next 6–9 months.
IDEAS WORTH REMEMBERING
5 ideasAgents are shifting AI from “chat” to “do work across tools.”
Sottiaux argues the key change is maturation: agents can now reliably execute multi-step tasks using computer/browser control and many integrations, so non-technical users won’t need to babysit workflows as much.
The mainstream workflow is “run overnight, approve in the morning.”
The envisioned near-term product pattern is a dashboard where agents complete tasks asynchronously and humans review/approve high-impact actions, compressing a day’s admin into a short decision loop.
Safety scales autonomy: use verification layers, not blind trust.
“Auto Review” adds a second agent to check the first agent’s actions for low-risk behavior, making longer autonomous runs more feasible—especially when sensitive inbox/calendar data is involved.
For personalization, give examples—not abstract rules.
Instead of writing a “tone of voice” description, provide real artifacts (newsletters, messages, transcripts) so the agent can infer style; similarly, keep project-specific folders and notes it can reference.
Expect cloud-based agent memory to replace device-bound setups.
Local-file workflows work today, but he predicts near-term migration to cloud-hosted memory/files so the same agent identity follows you across laptop/phone/travel without fragmentation.
WORDS WORTH SAVING
5 quotesThere is going to be dramatic change. Everyone is going to get their own little personal assistant on their computer that allows you to do more than whatever was possible three months ago, six months ago.
— Thibault Sottiaux
Today you have to actively prompt and sort of, like, you know, be creative about what you're, what you're asking... You know, in the future, that will not be true.
— Thibault Sottiaux
So you have the main agent doing actions, and then you have a second agent that is verifying all the actions of the first agent... and verifying that they are not doing anything that could be harmful to you and, like, are low risk.
— Thibault Sottiaux
At the end of the day, humans remain responsible.
— Thibault Sottiaux
It will, it will, it will be about engaging, you know, as your authentic self in a conversation and then getting the appropriate help in the right, at the right time.
— Thibault Sottiaux
High quality AI-generated summary created from speaker-labeled transcript.
