Tool of the Week: The model that didn't ship
This week the most anticipated AI model of the year was supposed to be the story. Instead, the story is that it isn't here.
Google's Gemini 3.5 Pro — announced back in May, positioned as the model to retake the lead — is months behind schedule. Bloomberg reported its coding ability came in short of what Google's own team expected, and the release slipped. On Thursday, Alphabet's stock fell about 4% on the news. A delay, not a failure, not a scandal — a delay — knocked tens of billions of dollars off one of the most valuable companies on earth in a single afternoon.
If a delay can do that to Google, it's worth asking what waiting is doing to you.
Here's the pattern I want you to notice. There is always a better model three months out. There was before Gemini 3.5, there will be after it. The demos are always stunning. The benchmarks always leapfrog. And there is a very natural, very expensive instinct that comes with it: I'll wait until the good one is out, then I'll set up the AI thing for my business.
That instinct feels prudent. It is actually how you never start.
Because the models that are already shipping — the ones you can log into this afternoon — already do far more than almost any small business is using. Claude, GPT, Grok, and Google's own current Gemini have all shipped capable versions in the last few weeks. The gap between "the best model that exists" and "the best model you've actually put to work" is enormous, and it has nothing to do with the release calendar.
The uncomfortable math: most owners are running maybe 10% of what today's tools can do, while telling themselves they're waiting for the tools to get better. The tool isn't the bottleneck. The setup is.
I'll say this as someone who builds with these things every day rather than reading about them: I don't wait for the next model. I run my operation on what's available right now, and I've never once been held back by the model being not-smart-enough. I've been held back by not having wired it in yet. Every time.
What to actually do, at any size:
Pick a task, not a model. Don't ask "which AI is best." Ask "what's one hour of my week I'd hand to a capable assistant today." Then use whatever's already in front of you to do that one thing. The model is more than good enough.
Treat the release calendar as noise. New model announcements are for labs and investors. For you, the question is never "is this the best model," it's "does the model I already have do this job." It almost always does.
Ship the 80% version now. The perfect setup on next quarter's model is worth less than a rough setup working this week. You can always upgrade the model later — it's usually a one-line change. You can't get back the weeks you spent waiting.
Let the giants chase the leaderboard. Google losing 4% over a delay is a big-company problem. Your advantage as a small operator is that you don't need the frontier — you need one useful thing turned on. That's available today, and it was available last quarter too.
Who this is for: anyone who's been meaning to "get around to the AI stuff" once things settle down. They won't settle down. There is no version of the future where the models stop improving and it finally feels safe to start. The owners pulling ahead aren't the ones with the best model. They're the ones who started with a good-enough one while everyone else waited.
The question to sit with this week: what's one task you've been postponing until the AI got better — and would today's tools actually already handle it if you just set it up?
Quick Hits
A delay wiped 4% off Google — and that's the real headline. Alphabet shares sank about 4% Thursday after Bloomberg reported Gemini 3.5 Pro is months behind schedule, with its coding ability short of internal expectations. Why it matters: the market just priced a delay as a serious blow, which tells you how fast the AI race moves — being three months late is now materially damaging. If a schedule slip hurts Google that much, the flip side is true for you: moving now, even imperfectly, is worth more than you think.
The "best model" crown moved again — this time away from Google. The same reporting notes rivals like OpenAI and Meta have recently shipped models that outpace Google's current offerings at generating code. Why it matters: whoever's "best" changes almost monthly. That's the whole case against welding your business to one model. Use tools that let you swap the engine, and don't build a strategy around today's leaderboard — it'll be different by the next issue.
Meanwhile, the cheap models quietly got good enough. Google's current-generation Gemini 3.1 Flash Lite shipped at roughly one-eighth the cost of its Pro model. Why it matters: while everyone watches the flagship race, the real small-business story is at the bottom of the price list. For categorizing an email, drafting a reply, or pulling a field out of a document, the cheapest tier is already plenty — and it costs almost nothing to run. You're probably overpaying for capability you don't use.
Prompt of the Week: The Ship-Now Audit
This week's theme is "stop waiting." So here's the 15-minute exercise that turns it into action. Paste this:
Act as a pragmatic operations advisor. I keep postponing setting up AI
for parts of my business because I'm waiting for the tools to get better.
Interview me one question at a time to cut through that.
First, help me list the specific tasks I've been meaning to hand off to
AI but haven't.
Then, for each task, tell me plainly: can a capable AI model that is
already available today do this to a "good enough to be useful" standard
— yes or no? Don't hedge. If yes, say so.
For every "yes," give me:
- The simplest way to start doing it this week, with tools I likely
already have
- What "good enough" looks like so I don't over-engineer it
- The one thing I'd have to do myself to set it up
Finish with the single task from my list that would save me the most
time for the least setup effort — the one I should turn on first.Most people run this convinced half their list needs "better AI," and come out realizing almost all of it works with what's already on their screen. The point isn't the prompt. It's noticing how long you've been waiting for permission that was never coming.
Like what you're reading? Forward it to someone who'd get value from it. And if you're curious what AI could actually do inside your business, book a free 15-minute audit — no pitch, just a look at where you're leaving time on the table.