🧠My honest take on GPT-5

The Interface Changes - A Mixed Bag

I genuinely feel like not having a giant list of models to choose from is a good direction.

Think about it, we went from this overwhelming dropdown with GPT-4, GPT-4 Turbo, GPT-4o, Claude, Gemini, and whatever else, to just "ChatGPT." It's cleaner, less decision paralysis, more focus on getting work done.

BUT deprecating all previous models overnight was a bad bad move.

Why this matters: Power users had workflows built around specific models. Maybe you knew GPT-4o was better for coding, GPT-4 Turbo for analysis, whatever. Overnight, all that institutional knowledge becomes useless. It's like your favorite restaurant suddenly removing everything from the menu except one dish - sure, maybe that dish is great, but you've lost all your go-to options that you knew worked for specific situations.

This is fundamentally different from how model transitions happened before. Usually you'd get overlap periods, gradual deprecation, choice in the transition. This felt more like "trust us, this is better" without letting users verify that claim themselves.

The Speed with Search - Actually Impressive

I really like the speed of GPT-5 with search enabled, the results are lightning fast. I was like "how are they finding all the relevant links and going through information so quickly?" I'll give them some points for that.

Why this is a big deal: Previous models with search were painfully slow. You'd ask a question, wait 10-15 seconds for it to search, then another 10-15 seconds to synthesize. It broke the conversational flow. You'd literally go make coffee while waiting.

GPT-5 with search feels almost instantaneous. It's hitting multiple sources, processing information, and giving you synthesized results in what feels like 2-3 seconds. That's a fundamental shift in how you can use AI with real-time information. Instead of "let me search this for you," it becomes "here's what I found" almost as fast as if it already knew the answer.

Strategic Brainstorming - Good, But With a Fatal Flaw

Then I tried using it for strategic brainstorming within projects → GPT-5 with thinking was far superior for these tasks compared to just GPT-5.

BUT it has a MASSIVE problem: I discovered that when I move from one project chat to another project chat, it completely loses it. There were two instances in 30 minutes where it was referring to a context I had given in a different chat in a different project.

Why this is unacceptable: This never happened with previous models. Each chat was isolated. You could work on Project A, then switch to Project B, and never worry about contamination. That's basic data hygiene.

I don't know if it's a rollout issue, but there's definitely some cross contamination of information happenin,g and that should be unacceptable. For anyone doing client work, handling sensitive information, or just wanting basic privacy between projects, this is a deal-breaker. It's like your email app accidentally includes recipients from previous emails in new messages.

Vibe Coding - Good Interface, Mediocre Results

Then I tried vibe coding a landing page, keeping it simple. I like the vibe coding interface, the speed, and the results were not that bad.

The interface improvement is real: it's more intuitive than previous coding interfaces, feels more collaborative, faster iteration cycles. The experience of working with it is genuinely better than earlier versions.

But then I compared the exact same prompt across other vibe coding tools, they were a lot better with following the design references I had attached.

This is the reality check that matters: GPT-5 might be technically better on paper, but in head-to-head comparisons on actual design work, it's losing to specialized tools. It's not understanding design references as well, not translating visual concepts as effectively.

The Bottom Line

Even though it's hitting all the benchmarks and some of the results I got from GPT-5 thinking were really good, as a power user the product feels like a regression.

Why a regression? Because the things that matter for actual work got worse:

• Lost model choice and control

• Interface improvements came with fundamental reliability problems

• Benchmarks don't capture the workflow disruption

It's like getting a faster car that randomly switches lanes without warning. Sure, it's technically better in some measurable ways, but you can't trust it for the drive to work.

That's my humble take after actually using GPT-5 for real work rather than just reading the benchmark claims.

👀 ICYMI

Must-Read AI Workflows

💻️ Which AI Coding Tool Should You Use? Let’s End the Tab-Hopping Confusion.
Stop juggling useless AI tools—find the one that actually fits your coding vibe and gets results. Whether you’re just starting or grinding daily, here’s your roadmap to smarter, faster coding.

📚 Prompting for Data Analysis: A Practical Framework
This new guide offers a structured approach to prompt engineering specifically for analysis, whether you’re working in e-commerce, SaaS, or client reporting. Includes super useful prompt templates.

AI Roundup

👶 Elon Has Launched “Baby Grok” — A Kid-Safe AI with Big Plans

Elon Musk’s xAI just dropped Baby Grok, a child-friendly version of its edgy chatbot — focused on safety, learning, and age-appropriate content.
It’s on waitlist for now, but signals a bold move into educational AI and responsible tech parenting.

🎭️ The Rise of AI Influencers—And Why You Shouldn’t Ignore It
The GEN platform is letting early users build branded AI personas from scratch.
Join the waitlist while it’s still low-key.