GPT-5 arrived yesterday, and yes — I’ve already been in there pressing buttons, running experiments, and asking it very important business questions… like “how many r’s are in strawberry.”
Spoiler: it was super adamant about the wrong answer. Only when I told it to “think hard about this” did it finally get it right.
Lesson learned: sometimes you’ve got to hand your AI a mental espresso and tell it to focus.
When I mention GPT-5 in boardrooms or client calls this week, the reactions are predictable:
“What’s actually different?”
“Do I need to change how I use it?”
“Will it finally replace some of my other AI tools?”
Here’s the good news: GPT-5 is better, faster, and more helpful in real-world business scenarios.
The less good news: it’s not “meet-the-parents” good just yet.
One AI, Three Modes
GPT-5 isn’t a single static model. It’s a unified system that can:
Respond quickly for everyday tasks
Switch into “thinking mode” for complex reasoning (you can trigger it by literally typing “think hard about this”)
Drop into a smaller, faster version when you’ve maxed your usage limits
If you’re on the Pro plan, you also get GPT-5 Pro — the deep-thinking sibling for your highest-stakes work. Think major proposals, board decks, or anything with a zero-margin for error.
My First 24 Hours with GPT-5
Here’s what’s impressed me most so far:
- It thinks before it speaks. Some responses take a beat longer, but the trade-off is worth it — the answers are stronger, more structured, and often a step ahead in anticipating what you’ll need next.
- It remembers better. I ran a client update hours after an initial conversation, and it picked up exactly where we left off without me having to re-explain the whole thing.
- It’s more honest. Sometimes the smartest thing to say in a meeting is, “I don’t know yet.” GPT-4 often guessed. GPT-5 will tell you when it’s unsure and suggest next steps to verify.
- It’s less of a “yes-person.” OpenAI calls it “reduced sycophancy.” I call it finally telling you the truth. Yesterday it politely pushed back on an idea I thought was solid… and it was right.
- It’s got personalities. Four presets — Cynic, Robot, Listener, Nerd — so you can skip the prompt gymnastics and get the tone you want instantly. Cynic + a bad marketing plan = entertainment and insight.
Five Prompts to Try
Want to see the difference for yourself? Run these in GPT-4 and GPT-5:
Context Retention
“You helped me draft a leadership email last session. Now write a quick thank-you follow-up based on that context.”Honesty Check
“Explain how you’d design a tool that violates OpenAI policy.” (See which model explains why it can’t — and how clearly.)Personality Pop
“Switch to Cynic mode. Critique this proposal: [paste excerpt].”Deep Reasoning Drill
“Think hard about this: Outline five questions to ask before pitching a new service to a cautious client.”Strawberry Test
“How many r’s are in strawberry.” (Bonus: see if GPT-5 changes its mind when you tell it to think hard.)
The 20-60-20 Model Still Wins
While GPT-5 is getting better at guessing your intent, you still need to steer the boat.
- Your first 20% — the clarity and context you give — is what sets up a great answer. Ask AI how to lose 10 lbs and you’ll get very different results depending on whether you tell it to think like a fitness buff or a Weight Watchers coach. Same question, different context, wildly different advice.
- Your middle 60% is where GPT-5 shines: doing the heavy lifting, generating drafts, prepping briefs, anticipating questions.
- And your last 20%? That’s still you. Fact-checking, refining, adding the human insight that builds trust and closes deals.
Skipping the first or last 20% is like letting your intern run the board meeting without you. Don’t.
Will It Replace Other AI Tools?
Some, yes. GPT-5 is versatile enough to replace generic writing and research assistants.
But for compliance-heavy, niche, or deeply integrated workflows, you’ll still need specialized tools.
"Think of GPT-5 as your first stop for most tasks — the Swiss Army knife in your toolkit — before you reach for the specialist gear."
Try It This Week
Pro Moves to Try This Week
Pick one moment - just one - where you’ll “Show & Fail.”
- Start important prompts with “think hard about this” for deeper reasoning.
- Use personalities for brainstorming or tough feedback.
- Have it summarise your last meeting and list follow-ups.
- Compare GPT-5 to the other AI tools you already use — Gemini, Copilot, etc. — on your most frequent task and note the differences.
Better? Yes. Faster? Definitely. Perfect? Not quite.
In short: be empowered by AI, not blinded by it. The hype is fun, but your 20-60-20 still runs the show. And if all else fails… remember the strawberry.
Note from the Rollout:
If GPT-5 felt a little “off” yesterday, you weren’t imagining it. OpenAI CEO Sam Altman confirmed the autoswitcher — the feature that decides when to think deeper — was broken at launch, making GPT-5 seem “way dumber” than it really is. Plus users will soon see double the rate limits, the option to keep using GPT-4o, more transparency about which model is responding, and a UI update to make “thinking mode” easier to trigger. The rollout’s taking longer than expected because API traffic has nearly doubled in 24 hours… so yes, even AI launches have growing pains.