🔥 Fire It In There Friday — 12/12/25

Code Red Was Real — and the Reality Is Messier Than the Headlines

There’s been no shortage of hype around Code Red, OpenAI, and the rapid escalation between OpenAI and Google over the past year.

Now that GPT-5.2 and Gemini 3 are both public and in active use, we’re finally past speculation.

But that doesn’t mean the picture is clean or settled.

---

Gemini 3: Past the Launch, Into Real Usage

With about a month of real-world exposure behind Gemini 3, the conversation shifts from launch speculation to practical impact.

And the impact is clear.

Gemini 3 isn’t just a model upgrade — it’s Google treating AI as an operating layer across the surfaces they already dominate:

• Search

• Workspace

• Android

• Chrome

• Cloud tooling

Its biggest strength isn’t one benchmark win.

It’s that intelligence shows up by default, deeply grounded in real-time information and multimodal context.

This is distribution power.

Google doesn’t need Gemini 3 to be the smartest model in isolation.

They need it to be everywhere people already work.

---

GPT-5.2: Strong on Paper, Rougher in Practice (So Far)

On paper, GPT-5.2 looks solid:

• Strong science and math benchmarks

• Deeper reasoning capabilities

• Clear focus on enterprise-grade reliability

• Continued investment in agents and tool-driven workflows

But here’s the honest, first-hand take — from someone who uses these tools every single day.

Even though GPT-5.2 was just released and I still have a lot more testing to do, my initial reaction is this:

GPT-5.2 feels a bit rough around the edges.

That surprised me.

My experience so far:

• GPT-5 felt like hopscotch — powerful, but you had to navigate carefully

• GPT-5.1 smoothed out many of those edges and felt more consistent

• GPT-5.2, in some responses, feels closer to that original hopscotch again

Not broken.

Not bad.

Just less polished in certain interactions.

Some responses require more steering.

Some flows feel slightly less predictable.

That doesn’t invalidate the benchmark gains — but it does reinforce something important:

Charts never tell the full story. Usage does.

---

Why This Isn’t Alarming (Yet)

This kind of roughness is common when platforms push deeper capability under pressure.

Code Red forced OpenAI to:

• Expand reasoning depth

• Push harder into complex domains like science and math

• Support more advanced, multi-step use cases

Those moves often introduce short-term friction.

Meanwhile, Gemini 3 feels smoother in many day-to-day interactions — not necessarily because it’s “smarter,” but because it’s embedded into workflows Google has refined for decades.

Polish comes from repetition at scale.

OpenAI is still tuning a general-purpose reasoning engine.

Google is refining a tightly controlled ecosystem.

---

What I’m Watching Next

I’m not calling winners.

I need more time with:

• Long-running workflows

• Agent orchestration

• Multi-document reasoning

• Daily professional use over weeks, not days

More testing is required, and I’ll keep everyone posted as patterns emerge.

Early impressions are useful — but only if we treat them as signals, not conclusions.

---

The Bigger Picture Still Holds

Despite the rough edges, the broader takeaway hasn’t changed:

• GPT-5.2 reflects OpenAI doubling down on reasoning depth

• Gemini 3 reflects Google doubling down on integration gravity

• This race is no longer about raw intelligence alone

• It’s about where intelligence lives and how smoothly it shows up

That’s a much harder problem than winning a benchmark chart.

---

Closing Thought

AI is officially past the demo phase.

Now we’re in the fit-and-finish phase — where small rough edges matter more than big claims.

GPT-5.2 is promising, but not perfect.

Gemini 3 is polished, but opinionated.

And the most honest answer right now?

We’re still learning how these systems behave in the real world.

Fire it in there.