Your electric bill is the one bill that could actually pay you back. A small but growing group of homeowners is charging batteries when power is cheap and selling it back at peak prices, with AI handling the timing. Here’s how ordinary people sell electricity with AI, and what it takes to join them.
The new way to make money from your power bill
Electricity is a weird commodity. Unlike corn or gasoline, it can’t be stored in huge amounts, so the grid balances supply and demand in real time. That makes prices swing hard: wholesale electricity can hit ten times the average in a single hour, and on very sunny or windy days it can even go negative, which means the utility pays you to use it.
For decades, only professional traders could profit from those swings. Not anymore. Home batteries, rooftop solar, and dynamic pricing tariffs have opened the door to regular homeowners. The play is simple: buy power when it’s cheap, store it in a battery, and sell it back a few hours later when prices spike.
The AI part is the timing. A script or agent watches the prices, decides the best moment to charge and discharge, and executes without you touching anything. Bloomberg documented this trend on September 4 in an article covering real households in the UK, Australia, and the US, and the numbers are surprisingly concrete.
Real people doing it
| Person | Where | Setup | Money made |
|---|---|---|---|
| Andrew Austin | Shropshire, UK | 200 kWh home battery, automated trades on Octopus Energy prices | ~£400/month in summer, £300 in winter, after household use |
| Aaron Wilkes | Kent, UK | Solar + batteries + 2 EVs run by an OpenAI Codex agent | Bills mostly covered; direct exports under £10/month |
| Mark Purcell | Queensland, Australia | Raspberry Pi trading hourly Amber Electric prices | Up to AU$200 in a single night, while sleeping |
| Greg Robinson | Arizona, US | FranklinWH battery in Ava Community Energy’s virtual power plant | $1,055 in December (signup + one sell event) |
Notice the range. Austin treats it like a side income, earning hundreds a month in summer. Wilkes mostly just covers his energy bills. Robinson got one big December payout and nothing since. Same idea, very different outcomes, and the difference is market rules and setup.
How the AI part actually works
Everyone who wants to sell electricity with AI today relies on the same loop, even if the tools look different.
- A price feed. Your utility publishes rates in advance. Octopus Energy in the UK publishes prices for every 30-minute period, and Amber Electric in Australia posts hourly prices. That feed is the brain’s input.
- A decision engine. A script, a Raspberry Pi, or an AI agent compares upcoming cheap and expensive windows against your household’s usage. It decides when to charge the battery and when to hold.
- An execution layer. The system controls your devices: discharging the battery at 7 p.m. when dinner cooking and prices peak, or pausing car charging at the worst hour.
Austin’s computer downloads Octopus’s prices every day and trades on its own. Wilkes, a train driver with no coding training, gave OpenAI’s Codex access to his device accounts and utility data. His words: “the devices came alive with an AI agent inside.” Purcell programmed a simple Raspberry Pi against Amber’s hourly rates.
Why bother with AI at all? A dumb timer could charge at night and sell every evening, but prices don’t follow a schedule. They move with the weather, the demand, and the occasional grid hiccup. An agent adapts daily: on a windy day it might skip charging entirely because prices barely move, and before a heat wave it may hold extra capacity for the spike. That adaptability is the difference between optimizing and guessing.
There’s a zero-code path too. Greg Robinson doesn’t run any scripts. His battery sits in a “virtual power plant,” where a utility combines thousands of home batteries into one grid resource. The utility runs the intelligence, and Robinson just chooses how much of his battery to make available.
What you need to get started
To sell electricity with AI profitably, you need four things, and the first three are non-negotiable.
- A battery (or solar plus battery). Solar makes daytime power free; the battery is what lets you time-shift it. Austin charges his 200 kWh battery from cheap grid power alone.
- A dynamic pricing tariff. If your utility charges one flat rate all day, there’s nothing to arbitrage. These tariffs exist in parts of the US, most of Europe, and Australia.
- A smart meter. It doesn’t just measure usage, it tells you the price at the moment of consumption. Countries like Sweden, Italy, Spain, and Finland have near-universal adoption.
- Automation. A script, an AI agent, or a VPP program that handles the selling for you.
The last one is where beginners usually land. Virtual power plant programs let you participate without writing any code, which is probably the smartest entry point for most readers.
The risks (and the honest catch)
Let’s not sugarcoat it. Before you rush to sell electricity with AI, know what can go wrong.
Agents make mistakes. Wilkes once watched prices climb while his car kept charging, because the AI had lost access to the vehicle and couldn’t stop it. A price cap from his utility capped the damage at about £1 per kWh, which is the kind of safety net you want to confirm exists before you start.
Grid costs can sink the plan. Wilkes wanted bigger batteries to trade more, and his grid operator quoted him tens of thousands of dollars in upgrades. That killed the economics instantly.
Regulation is the biggest gate. Only Australia, most of Europe, and some US states let households sell at hourly or half-hourly prices. In most of the world, this entire article doesn’t apply yet. And even where it works, payouts vary: Wilkes’s exports rarely exceeded £10 a month, while Austin’s summer months hit £400. Your mileage will depend on your market, your hardware, and your luck with the weather.
Is it worth it for you?
Honest take: this is real, it’s documented, and it’s not hype. But it’s also not a universal money hack. The deciding factor is where you live.
Start with your utility’s website. Look for two things: time-of-use rates that change by the hour, and any virtual power plant program. If your utility offers either, the math is worth running. If it offers neither, no amount of AI will help you arbitrage a flat rate.
The hardware costs real money too. A home battery runs thousands of dollars, so the payback window only makes sense if you already have solar, a big battery, or a VPP that pays for participation. For most people, the smarter move is joining a VPP before buying anything.
Takeaway
Check your utility’s export rates this week. That single lookup tells you whether your market supports selling power back at dynamic prices at all. If it does, a battery plus automation is a genuine side income, and to sell electricity with AI successfully you don’t need to understand the code, you need the right tariff.
Curious when automation genuinely pays off? Our MIT research breakdown covers the math. And if you want to see how AI agents like the ones running these trades are built, the Warp self-improving agents guide is a solid primer.
