Not answer their questions. Spend their money.

The project is called Hatch, and it is not another chatbot. It is a persistent AI agent that remembers you across days and weeks, works toward goals in the background, and takes real actions on your behalf. Booking services. Managing purchases. Rescheduling meetings. Completing a checkout without being asked twice.

One caveat before we go further, because it matters: none of this is confirmed by Meta. Everything below comes from reporting by The Information, the Financial Times, and Business Insider. So treat it as a strong signal, not a spec sheet.

The signal is still worth acting on. Here is why, and what I would do about it if I ran a business in Cyprus.

What Hatch actually does

Most AI assistants forget you the moment a session ends. Hatch is built to do the opposite.

It keeps context. Tell it you need a caterer for an event in three weeks, and it is meant to remember, follow up, and track progress without a nudge from you.

It works on goals. You describe an outcome, and the agent takes the multi-step path to get there, checking back only when it needs input or when something changes.

It connects to your services. Reportedly email, calendar, and third-party apps. Not copy-paste integration. The agent is designed to read your inbox, find the threads that matter, draft the replies, and move the meetings.

It uses a browser and a computer on its own. To train it, Meta reportedly built sandboxes that simulate real services like DoorDash, Etsy, Reddit, Yelp, and Outlook. That last capability is the one everyone keeps talking about. An AI that can buy things, not just recommend them.

And it lives where people already are. Not as some app you have to remember to open, but inside Instagram and WhatsApp, in front of more than two billion daily users.

That last detail is the whole story. Hold onto it.

Who is actually building it

Here is the part that should make you stop scrolling.

During development, Hatch is reportedly running on Anthropic’s Claude, not on Meta’s own models. Claude Opus 4.6 and Claude Sonnet 4.6, specifically.

Sit with that for a second. Meta, a company spending well over a hundred billion dollars on AI this year, is reportedly paying a direct competitor to build the backbone of its most ambitious consumer product. One report put Meta’s projected spend on Anthropic’s models at up to ten billion dollars a year.

Meta plans to swap in its own model before launch. The naming in the reporting is messy. Some outlets say Muse Spark, others point to an in-house model codenamed Watermelon. Either way, the plan is the same: build it on the best available model now, replace it later.

You can read that as an accident of deadlines. I read it as the most honest benchmark in the industry. When the company competing hardest wants the job done well, it reaches for Claude.

Meta is not alone in the race. Google is reportedly testing its own persistent agent, first codenamed Remy, since rebranded toward Gemini Spark. OpenAI is pushing its Operator ecosystem in the same direction. Anthropic already shipped Claude Cowork. Five big players, five agents, one promise: take the repetitive work off your plate.

Hatch stands out for one reason. Distribution. Nobody else can drop an agent into two billion existing accounts overnight.

What it costs

Official pricing has not been confirmed. Reporting points to tiered subscriptions, with a premium plan reaching around $199.99 a month for the full set of goals, connectors, and computer use.

If that holds, Hatch arrives as a prosumer and small-business product, not a free mass-market toy. It would also be Meta’s first paid consumer product, from a company that has monetised through ads for its entire life.

For any business already paying for a virtual assistant, scheduling software, or basic automation, that number will feel familiar. At $200 a month, Hatch would cost less than four hours of outsourced admin.

What it means for businesses in Cyprus

The practical question is not whether this technology is impressive. It clearly is. The question is how fast it changes what your customers and your staff expect.

Start with the customer side.

If a customer’s agent can research, compare, and complete a purchase on its own, the businesses it picks will be the ones built for a machine to read. Clear pricing. Machine-readable availability. Fast, structured answers to queries.

The businesses that rely on a long back-and-forth WhatsApp thread to close a sale? They will find that thread quietly skipped. The agent will not haggle over three days. It will choose the option it can actually complete.

That is the part Cyprus operators should feel most directly. Hospitality, real estate, legal and accounting firms, retail. A lot of local selling still happens in exactly the informal, conversational way an agent is worst at using.

Now the internal side.

A persistent agent that manages calendars, drafts emails, tracks supplier conversations, and handles recurring admin is a description of what a two-person business does by hand today. Or pays a part-time coordinator to do.

The disruption here is not dramatic. It is slow compression. Scheduling. Follow-ups. Research. Purchase completion. Each one small on its own. Together, a real share of the overhead of running a small business.

The timing

There is no confirmed launch date. Reporting from late summer described a rollout in “the coming weeks,” with a consumer release aimed at late 2026.

So this is not a someday problem. Given the pressure from Google and OpenAI, the timeline is measured in months, not years. It may already be closer than that by the time you read this.

Final thoughts

Here is the move I would make now, before the tools ship broadly.

Take an afternoon. List the parts of your operation that could be handed to an agent, and the parts that genuinely need a human to decide. Then structure your public-facing information so an agent can actually use it. Clear prices. Real availability. Answers a machine can parse without a phone call.

The companies that run that review now will adopt these tools fast. The ones that wait will spend next year catching up to competitors who did not.

Persistent AI agents stopped being a research concept. They are being tested at scale by the largest technology company in the world, on the best model its biggest rival will sell it.

That changes the conversation. The only question left is whether your business is ready to be spoken to by a machine.