Head of ChatGPT & Codex: agents for normal people are HERE

📌 Try the Liberty 5 Pro series by @SoundcoreAudio FREE for 30 days via this link: https://soundcore.tech/D1204_DTC_Listing_Silicon_Valley_Girl_0_505535 Thibault Sottiaux is the head of ChatGPT and Codex at OpenAI. He told me that in a few months, people who don't use AI at all will get the same benefits as those who've spent two years figuring it out. So what's actually going to be your edge in a market this competitive? Thibault opened up and shared a few of his secrets — he showed me how his own agents work and how anyone can set them up. We covered the must-have files everyone should create, the one file you should never write yourself, and the new skill that replaces prompting entirely. If you work with a computer for a living, this is the one to watch this week. *Timestamps:* 00:00 — The change nobody is ready for 00:47 — How knowledge work changes tomorrow 03:00 — The agentic workflow breakthrough 04:18 — The must-have files everyone needs 05:54 — The one file you should NOT write yourself 07:08 — Use agents vs. don't — the productivity gap 07:50 — The trap of optimizing everything 11:47 — Vibe coding: when you still need an engineer 13:38 — The future of software engineering 15:07 — A workflow you should deploy today 16:30 — Live demo: agent runs my inbox + plans a trip 21:18 — "Where am I wasting my time?" 22:48 — What Thibault personally uses Codex for 24:53 — Live demo: agent pulls my LinkedIn analytics 25:55 — He quit his PhD after 2 weeks — would he do it today 27:48 — The "personal tailor" model of AI 29:04 — The real skill that replaces prompting *Links:* 📩 Follow my Newsletter: https://siliconvalleygirl.beehiiv.com/subscribe?utm_source=youtube&utm_medium=video&utm_campaign=futureproof-sub&utm_content=Thibault-Sottiaux 🔗 My Instagram: https://www.instagram.com/siliconvalleygirl/ 📌 My Companies & Products: https://Marinamogilko.co

Thibault SottiauxguestMarina Mogilkohost

May 22, 202631mWatch on YouTube ↗

CHAPTERS

0:00 – 0:47
Personal assistants for everyone: the “change nobody is ready for”
Thibault Sottiaux predicts a dramatic shift where everyone gets a reliable personal assistant on their computer. The key change: people won’t need to learn prompting tricks to benefit—AI will deliver value by default.
0:47 – 3:00
How agents will reshape knowledge work day-to-day
They discuss how agents will impact marketers and other knowledge workers by automating routine research, email triage, and prospecting. Scheduling agent tasks (like cron-style runs) becomes a mainstream feature inside apps.
3:00 – 4:18
The agentic workflow breakthrough: reliability + tool access
Thibault explains that the technology has matured: agents can operate reliably over longer horizons and use many tools (browser/computer use plus numerous integrations). This eliminates the need for users to “babysit” or configure technical details.
4:18 – 5:54
Must-have agent files: tone, projects, and examples (not explanations)
Marina asks how to organize data so agents can be effective. Thibault recommends keeping tidy project folders and using examples to teach tone of voice rather than trying to describe it explicitly.
5:54 – 7:08
The one file you shouldn’t write: ‘tone of voice’ as samples, not rules
Thibault emphasizes a counterintuitive point: don’t attempt to author a detailed tone guide. Agents learn better from representative writing samples across contexts (professional vs. personal).
7:08 – 7:50
Use agents vs don’t: productivity gains—and the responsibility tradeoff
They contrast people who adopt agents with those who don’t, noting adopters will unlock more throughput and tackle deferred tasks. Marina raises a responsibility dilemma: automation increases output, but humans must still verify and own results.
7:50 – 11:47
The trap of optimizing everything: capability curve and burnout risk
Marina describes feeling overwhelmed by trying to optimize too much with agents. Thibault warns about pushing beyond today’s reliability frontier—useful for discovery, but it can create frustration until models improve.
11:47 – 13:38
Vibe coding: when you still need an engineer
They define when vibe coding is sufficient (small prototypes, personal tools) versus when an experienced engineer matters (architecture, scalability, maintainability). Thibault expects agents to increasingly handle long-term structure, but not fully yet.
13:38 – 15:07
Future of software engineering: more software, not less demand
Thibault argues AI will trigger an explosion in apps and infrastructure because it lowers the cost of building. Even with AI writing more code, demand for technical talent persists due to endless new problems and products to create.
15:07 – 24:53
A workflow to deploy today: daily briefs, inbox triage, and trip planning
Thibault demonstrates how to set up agentic threads that run in parallel—summarizing updates, scanning inbox threads, preparing replies, cleaning Gmail filters, and planning trips using calendar availability. The focus is turning recurring “busywork” into repeatable agent tasks.
24:53 – 27:48
Computer-use demo: agent downloads LinkedIn analytics + turns workflows into skills
They run a computer-use agent that navigates LinkedIn, exports analytics, and produces a spreadsheet, then refine the request for “impressions per post.” Thibault explains “skills” that package a workflow so it can be rerun on a schedule.
27:48
Bigger life choices and the ‘personal tailor’ model: the skill replacing prompting
Thibault reflects on dropping out of a PhD, following energy/instincts, and his path through Google/DeepMind to OpenAI. He frames AI’s future as a “personal tailor” that understands you through conversation—where authenticity and asking the right questions matter more than prompt craft.
Autonomy with safeguards: Auto Review and “approve what it did overnight”
They explore the emerging “dashboard” model where agents do work asynchronously and humans approve outcomes. Thibault highlights Auto Review: a second agent verifies actions for safety, enabling more autonomous operation with sensitive data.
Local vs cloud memory: the upcoming shift in how agents store context
They discuss the friction of local file setups when moving across devices. Thibault predicts agent memory and file management will move to the cloud soon, reducing fragmentation between laptop/phone/work machines.
Live build: fixing a voice feature with APIs—and why ‘technical’ still matters
They review a content repurposing app Marina generated and troubleshoot missing voice dictation. Thibault shows how to prompt for integrating Speech-to-Text via OpenAI APIs, highlighting that today’s gaps still require some technical intuition (API keys, docs).