Aug 18, 2025 | Issue 12

The long-awaited iteration of the model that started the public love/hate affair with AI landed last week. And (predictably) the reaction is still love/hate.

What I’m seeing:

• Contextual control is now table stakes. GPT-5 brings fine-grained personality modes and easier context engineering. If you rely on tone and precision, mastering these is no longer optional.
• Model deprecation shattered trust. Legacy models like GPT-4o and the o-series vanished overnight. Workflows (and yes, botfriends!) broke when the new router UI took over.
• GPT-OSS lands but barely ripples. The open-weight gpt-oss-120B and 20B are here under Apache 2.0, but adoption is tiny. At only ~14 GB with 16 GB RAM, you can run it locally — like having a sandboxed o4-mini for that old PC in the corner.
• Prompt fundamentals endure. “Think step-by-step,” chaining, and decomposed instructions still work as well as they ever did.
• Pricing and throttling outrank features. The biggest talk isn’t capability — it’s invisible throttles and cost changes, especially for Claude users.
• Prompt versioning is now mission-critical. As prompts get more complex, managing them like source code — versioned, tracked, and documented — is the only way to scale without chaos.

Why it matters:

Before, you could ignore some of the “best practice” nagging and still get by. Now? Not if you’re serious.

• If you’re not managing prompts, you’re not serious about AI.
• If you’re not tuning settings, you’re blunting the tool.
• If you haven’t grabbed gpt-oss, you’re betting your workflows on OpenAI’s next unpredictable decision.

🧠 Strategic Human Prompt

What’s one GPT-5 change that genuinely improves your workflow? Or… What’s one GPT-4 era habit you’re keeping, no matter what?

➖ Suggested Subtraction

Follow OpenAI’s lead They killed every previous model and rewired the user experience in one day. If a $300–$500 B company can do that, what’s the long-delayed change you’ve been too cautious to make?

Find it. Accelerate it. Rip the band-aid off, then explain why. Make the resistance look backward-facing, then circle back for the stragglers.

Analogy

tbd

♬ Closing Notes

GPT-5 isn’t a rewrite. It’s an incremental but important step. Reasoning, multimodality, and safety improved, but the fundamentals didn’t change. The most durable workflows:

1. Treat prompts as first-class code
2. Version prompts and context like software
3. Scale with discipline (and backups)

Want to boost your organization’s immunity to these kinds of disruptions going forward? Hit reply and let’s talk.

Until next time,

Read on Substack