Meta builds personalised agentic AI assistant on Muse Spark model

TL;DR:

  • Meta is building a personalised agentic AI assistant for its billions of users, the Financial Times has reported. The assistant is powered by Meta’s new Muse Spark AI model and is being tested internally under the codename “Hatch”.
  • Separately, The Information reports Meta plans to integrate an agentic shopping tool into Instagram, targeting a launch before Q4 2026. Meta has not commented.
  • Resultsense view: Meta now joins OpenAI (OpenClaw), Anthropic (Claude finance agents) and an in-development Google equivalent in pushing agentic AI into the consumer surface this quarter. UK consumers and SME advertisers using Instagram should expect commerce, support and recommendation flows to change materially before Christmas trading.

The disclosures come as Meta faces investor scrutiny over escalating AI spending. Last month the company raised its annual capital-spending forecast, signalling plans to pour billions more into AI infrastructure even as it confronts potential losses from a global youth backlash against social media.

What is being built

The agentic assistant under codename “Hatch” is positioned by reporting as a Meta equivalent to OpenAI’s OpenClaw — connecting hardware and software tools, learning from data with much less human intervention than a chatbot. The internal goal, per The Information, is to complete internal testing by the end of June 2026.

Muse Spark, the underlying model, is a new Meta family that sits in the assistant role rather than the foundation-research role Llama traditionally has held. The split mirrors OpenAI’s distinction between GPT-5.5 Instant (consumer default) and the smaller agentic models powering the OpenClaw stack.

The Instagram angle

The agentic shopping tool is the more commercially material near-term move. Instagram’s commerce funnel — discovery, save, message-the-seller, checkout — is one of the most-trafficked consumer-purchase paths in the UK. Putting an agent inside that funnel changes how products are found, compared and bought.

For UK SME retailers and creators, three direct effects become possible. First, AI-led product comparison may displace creator-led recommendation in some categories. Second, agentic checkout reduces purchase friction but increases regret risk — UK consumer-protection rules (Consumer Rights Act, distance-selling regulations) will need clearer guidance on agent-mediated purchases. Third, ad pricing dynamics shift: when an agent picks a product the user buys, the attribution conversation is harder for advertisers.

How this fits with the wider agentic push

The same week, Anthropic shipped ten finance agents and Microsoft 365 add-ins, OpenAI launched GPT-5.5 Instant with stronger memory-source controls, and FIS announced a Claude-powered Financial Crimes AI Agent. Across consumer, enterprise and finance, the operative move from frontier labs is the same: stop selling chat, start selling agents.

The strategic question for UK businesses is which surface to integrate against. Instagram is where UK consumers shop. Microsoft 365 is where they work. ChatGPT is where they ask. Each is now an emerging agentic platform with different governance affordances and different regulatory exposure.

UK relevance

UK businesses should plan for three near-term checkpoints. First, the Q3 2026 Instagram-Hatch launch (or its slip), which will signal whether Meta can ship in line with The Information’s timeline. Second, ICO and ASA guidance on agentic-commerce attribution and consumer disclosure, both of which are overdue. Third, whether UK CMA’s recent merger and conduct review of agentic AI in adjacent markets shapes how Meta integrates Hatch with Instagram-side seller tools.

Looking forward

If Meta hits a Q3 launch, the holiday-trading 2026 season is the first market test of agent-mediated commerce at consumer scale in the UK. UK retailers should review their Instagram seller integrations now — content optimised for human discovery and content optimised for agent retrieval are different problems and will diverge in best practice through 2026.

Share this article