GPT-5 Just Proved Why You Must Own Your AI

Friends,

As I'm relaxing in the Algarve I wanted to take this opportunity to focus on what last week's rollout of ChatGPT-5 teaches us about alignment with AI. This is a new format I've been playing with that I think could complement the weekly news breakdown you're used to seeing and I'd love to know what you think.

TL;DR

A clear signal for shaping your AI strategy:

• What happened: GPT-5 launched with auto-switching models, removed user favourites, and caused disruption before rolling back changes.

• Why it matters: If you don’t own your AI stack, you’re at the mercy of sudden platform decisions.

• Your move: Own your AI OS, move to APIs, build fallback capacity, and align AI to your business — not the other way round.

What happened

OpenAI’s GPT-5 rollout in early August 2025 was messy. A new real-time model router auto-switched between "fast" and "thinking" modes, but for many users this caused fluctuating performance, inconsistent tone, and sudden removal of popular models like GPT-4o without warning. The public backlash was fast. CEO Sam Altman reinstated GPT-4o for paying users, adjusted rate limits, and added manual mode selection to restore control. He also apologised for misleading charts shown in the launch livestream, calling it "chart crime". Read the TechCrunch piece for the full story.

Why it matters

• Dependence on platform decisions is risky. When you don’t own your AI stack, providers can change models, limits, or interfaces without notice.

• Owning your AI tech stack equals alignment. If the provider can flip a switch and alter your tools, you lose control.

• Control protects cost-effectiveness. AI feels cheap now because providers want to lock you in. Once you’re dependent, they can raise costs or introduce constraints.

• Rollout pain was greatest at the product layer. ChatGPT UI users got the full impact. API users experienced less disruption, though even Custom GPT creators were affected.

• AGI hype vs. reality. GPT-5 was billed as a step toward AGI, but the rollout felt incremental and poorly executed.

Your AI Optimistic Playbook

1. Decide who is going to own your AI Operating System
   Make clear who in your organisation is responsible for architecting and governing your AI stack. This is no longer optional.

2. Prioritise moving Custom GPTs to the Assistants API
   Start building directly on the Assistants API rather than relying solely on the ChatGPT UI. It gives you more control and flexibility.

3. Build fallback options
   Test alternative providers or local models so you can switch quickly if a platform changes terms, pricing, or model performance.

4. Monitor and test continuously
   Track how models behave, cost, and perform over time. Run A/B tests between models and across vendors to maintain quality.

5. Align AI tools with your values and workflows
   Your AI shouldn’t just be convenient — it should be tailored to your business logic, compliance needs, and strategic goals.

Work with me

Now accepting AI strategy workshops for Q3/Q4 2025

Owning Your AI: Building Alignment Through Your Tech Stack - I’m delivering 90-minute executive sessions for senior teams on designing and implementing an AI Operating System, including governance, adoption, and alignment frameworks. Available across the UK, Europe, and globally, in person or virtually.

To book for your board, team away day, or keynote event, fill in this form.

Stay optimistic,

Hugo & Ben

AI Night School & The AI Optimist