Two of the biggest AI companies in the world made announcements this September that have nothing to do with each other on the surface. Meta launched a personal AI agent that video calls you. Anthropic quietly dropped the price of its best coding model by more than half.
Put them side by side and they are saying the exact same thing from opposite directions. Intelligence is not just getting better. It is getting cheap enough to actually reach the rest of us, Kampala included.
That is the part worth paying attention to. Not the demo. The price tag.
Meta built an agent people actually wanted
Muse launched on September 8. It is Meta's personal AI agent, built on a model called Muse Spark, and it does the thing every previous AI assistant promised and mostly failed to deliver: it actually runs errands.
Connect it to your calendar, your email, your shopping accounts, and Muse will book the appointment, place the order, and manage the back and forth. It has its own email address, so you can loop it into a thread the way you would a human assistant. It works through more than 1,500 developer connectors, including Shopify, PayPal, Best Buy, Walmart, Expedia and GitHub. On Meta's smart glasses, you can talk to it hands free to log a meal, guide a workout, or add something to your cart. And if you want a face to talk to, Muse has a real time video avatar you can video call like a person.
Meta gave away 100 million tokens a week for free and still charges $20 and $100 a month for heavier use. It didn't matter. Muse hit number one on both the Apple App Store and Google Play in the United States, pulling in 2.5 million downloads in thirteen days, faster than ChatGPT managed at launch. Meta's stock jumped nearly 27% in the two weeks after, including its single biggest one day gain in years.
That is not a curiosity. That is proof that ordinary people, not just developers, will adopt an AI agent the moment it is genuinely useful and easy to reach.
Anthropic's real headline was the invoice, not the benchmark
Three weeks later, Anthropic released Claude Opus 5.5, its flagship coding and reasoning model. The benchmark numbers are strong on their own. It scored 66.4% on Terminal-Bench 4.0, a test of real agentic coding tasks, up from 52.3% on the previous Opus. On a knowledge work benchmark spanning 44 occupations, it posted the highest score Anthropic has recorded.
But the number that actually changes who can use it is the price. Opus 5.5 runs at $24 per million tokens combined, against $60 for Anthropic's own larger Fable 5.1 model released three weeks earlier. That is a 60% cut, and Opus 5.5 still beats Fable 5.1 on most of the coding benchmarks that matter for real work. In one published test, it audited and repaired a 200,000 line codebase in under three hours. The previous Opus needed more than twenty, burning two and a half times the tokens to do it.
Translate that into a currency a small software team understands. A job that used to take most of a working week now takes an afternoon, and costs a fraction as much per run.
Why the price is the part that reaches Uganda
A frontier model that only a well funded lab in San Francisco can afford is interesting to read about and useless to build with. A frontier model that costs 60% less and runs in an afternoon what used to take a week is a tool a two person software shop in Kampala can actually put into production.
It's the same reason mobile money mattered more than mobile banking apps ever did here. The technology existed either way. What changed lives was the price of using it dropping to something an ordinary person could afford on an ordinary week.
That is what happened twice in the same month, once in a consumer product with 2.5 million downloads, once in an API price sheet most people will never read. Both are the same signal. The gap between what Silicon Valley can build with AI and what a small team in East Africa can build with AI just got smaller, not because Africa caught up, but because the tools came down to meet us.
What this actually opens up
For a software or automation team here, a coding model that handles a five figure line count job in an afternoon at a fraction of last quarter's price means the backlog moves faster and the pricing conversation with a client changes shape entirely. Work that wasn't worth quoting six months ago is worth quoting now.
For a business watching Muse, the lesson isn't the avatar. It's that customers have just shown, at scale, that they will hand real tasks and real purchasing decisions to an AI agent the moment it works. That appetite doesn't stay confined to the US market it launched in. It sets the expectation everyone building customer facing AI here will be measured against next.
None of this is a five year prediction. It happened in the same three weeks.
Amazing times ahead, if you build for them
The frontier didn't just move forward this month. It got affordable enough that where you happen to be building from matters less than it did in August.
That window doesn't stay this wide open forever. The businesses that use it now are the ones who'll be ahead when it closes.
By Nasif K Balinda
