How much better can frontier models get?
My day-to-day involves sending emails, building decks, and writing fluffy marketing shit. It’s now so easy to connect Claude/GPT into a few systems and quickly realize significant productivity gains.
And Fable, Claude’s supposed God-model, while impressive, didn’t make my workflows all that much better. I feel like I’ve optimized all I can at this point.
Take this with a grain of salt. I’m not a software developer, biomedical engineer or mathematician. There are probably still significant gains to be made with high IQ work.
But if I were the frontier labs, which desperately need to improve profitability (for the sake of the world economy), I’d focus on getting this existing magic into the hands of as many people as possible vs. very expensively improving the models.
Yes, enterprises are adopting them en masse. But this is mostly being done at an individual level while CFOs are scrambling to reign in company spend.
It’s proving hard to figure out deployment that is safe and cost-effective, with minimal disruption to a large cross-functional operation.
So…where is the whitespace?
The Small Business Opportunity
The United States has more than 30M small businesses. Unlike enterprises, most operate without layers of systems, processes, and labor that require years of strategic planning to rearchitect around AI.
These companies manufacture components, distribute industrial products, install windows, supply building materials, fabricate metal, repair equipment, etc. Their value comes from moving physical goods, solving real customer problems, and building relationships that often take years to develop.
AI won’t fundamentally change what these companies do - which is probably why it isn’t on the radar for many of them. But there is significant opportunity for AI to improve how they operate.
Consider a company generating $4 million in annual revenue. It has a handful of office staff quoting projects, responding to customers, entering orders, coordinating schedules, and keeping paperwork moving. Most of this lives in spreadsheets, email inboxes, and the accumulated knowledge of employees who have been there for years.
Take something as simple as a request for quote. Today, an employee might:
Review the customer’s specifications
Search previous jobs for similar work
Gather pricing from suppliers
Build the proposal
Send it to the customer
Repeat the process when revisions come back
Do that dozens of times a week and you’re spending a lot on labor for fairly repeatable process.
Much of that workflow can happen automatically. An intelligence layer can:
Extract and organize the specifications
Surface similar historical projects
Pull relevant pricing
Generate a first draft of the proposal
Flag anything that requires human judgment
Hand the finished work to an employee for review
A process that previously took hours can take minutes.
This can repeat across the business. Customer emails can be drafted automatically. Meeting notes can populate the CRM. Purchase orders can be checked against historical pricing. Documentation can be generated as work progresses. Scheduling can adjust as priorities change.
None of these improvements move the needle that much individually, but collectively can change the economics of the business.
Margin Expansion + Growth
Suppose an estimator who previously completed ten quotes each day can now complete thirty.
Before AI, the business may have needed three estimators simply to keep up with demand. Maybe one of them wasn’t particularly good, but the company needed the extra capacity to fulfill demand.
With AI, the two stronger estimators are no longer buried in administrative work. They can handle significantly more volume, which means the business can eliminate unnecessary labor while still increasing the number of opportunities it pursues.
The economics could look something like this:
Before AI
$4M revenue
10 quotes per estimator per day
3 estimators
8% net margin
$320K annual profit
After AI
$5M revenue
30 quotes per estimator per day
2 estimators
12% net margin
$600K annual profit
This is illustrative but that’s 25% more revenue and 88% more profit.
The point isn’t that every business will eliminate a role. It’s that AI gives good employees significantly more capacity. That lets a company remove labor it only carried because of inefficient processes, while also taking on more work without immediately hiring again.
The Small Business Summary
Grow without adding headcount as existing employees handle more volume
Make its best employees more productive by removing repetitive administrative work
Reallocate time toward growth through more customer visits, follow-up, and account expansion (things that only a human can do)
Reinvest higher cash flow into salespeople, equipment, new locations, or acquisitions