The Fastest AI Answer Is One Finished Before You Ask
AI founder Neil Movva says the future is assistants that watch your inbox and line up your next move, which only works if computing gets cheap enough to waste.
Invest Like the Best with Patrick O'Shaughnessy · Neil Movva - Making AI 10x Cheaper - [Invest Like the Best, EP.488]Neil Movva, founder of a company that sells cheap computing for artificial intelligence, was explaining why he is building for AI helpers that work through long jobs on their own when he offered his version of speed.
“The best latency is no latency at all,” Movva said. “When you wake up in the morning, the work’s already been done overnight.”
By latency, he means the wait between asking an AI tool for something and getting an answer. The industry has spent the past two years trying to shrink that wait to make chatbots feel instant. Movva argues that race misses the point, because it keeps a person in the loop, prompting and waiting. While you wait, he said, you are the bottleneck that stops the agent from doing more work.
What he wants instead is background work on human time scales. Movva said, “You don't manage your colleagues every five minutes.” You give them a broad task and check back in a day or a week later. He believes people will soon work with AI the same way, handing off long jobs and returning only when there is something to review.
His company, Sale Research, is built around that bet. It sells access to open source AI models at the lowest possible price and rents out long running workspaces in the cloud where agents can keep working for hours, days, or weeks.
The dream he described is a phone assistant that runs all the time. It would read the day’s emails and text messages, building what Movva called “a much more encyclopedic view of your life” to figure out what you are likely to do next. When you open your phone, the next action would already be waiting.
That kind of help, Movva said, only works if intelligence gets so cheap that developers are willing to spend it “without any promise of return,” running endless quiet calculations in case one of them turns out to be useful.
Questions people ask
Do any agents work this way today?
Yes, in a limited way. Research tools can already spend 10 or 20 minutes scanning dozens of sources and return a report, and coding agents can work for an hour while you do something else. What they do not yet do well is run reliably for days, remember everything about you, and decide on their own what is worth doing next.
What is AI inference, and why does running it overnight cost anything?
Inference is the computing work of running a trained AI model to answer questions, as opposed to training it in the first place. Providers charge for it by the token, a small chunk of text, so a long background job that reads thousands of emails can burn millions of tokens before it ever helps you.
What has to be true about privacy for this to work?
It raises a tradeoff between helpfulness and privacy. A truly proactive assistant would need constant access to your emails, messages, calendar, and location, and that data would have to be stored and processed somewhere secure. Movva said users will have to trust that the machine is both reliable and private before this works.