Meta Forced to Unwind $2B Manus Agent Deal as New Study Exposes AI Coding Agents' Blind Spot

Meta has started dismantling its $2 billion acquisition of Manus, the general-purpose AI agent, after Beijing ordered the deal reversed, TechCrunch reported. The forced unwind reshuffles the agent market: Manus is one of the few commercially deployed general-purpose agents, and its uncertain ownership creates an opening for competitors while raising questions about how geopolitical friction will shape future agent M&A. No official statement from Meta or Manus has been released, but sources inside Meta confirmed that unwinding teams have been formed.

A new study covered by The Decoder reveals a systematic weakness in current AI coding agents: they reliably find the right file in a codebase but consistently miss the exact lines that need modification. The research tested multiple agent architectures and found that file-level accuracy is high while line-level precision drops sharply — a gap that can't be bridged by simply scaling context windows. Every major coding agent, from OpenAI Codex to Claude Code and Cursor, is affected. The finding redirects attention toward code-grounded reasoning architectures rather than bigger models.

Anthropic published research on turning Claude into a chemistry research agent, demonstrating autonomous planning and execution of multi-step synthesis workflows. Claude selects reagents, predicts outcomes, interprets spectroscopy data, and iterates on failures without human intervention. The work represents a template for how frontier models can be specialized as scientific discovery agents, moving beyond code generation into experimental chemistry.

The Anthropic Fable 5 shutdown story gained a new dimension: the Wall Street Journal reported that Amazon CEO Andy Jassy's direct conversations with US officials triggered the government's decision to force Anthropic to disable Fable 5 and Mythos 5 worldwide. The revelation, which scored 677 points and 497 comments on Hacker News, adds a layer of corporate politics to what was already the most significant AI governance event of the year. Amazon is Anthropic's largest investor and primary cloud provider.

Source-linked headlines

Meta reportedly moves to unwind $2B Manus deal after Beijing's demand

TechCrunch AI · June 13

Meta begins dismantling its $2 billion acquisition of Manus, the general-purpose AI agent, after Beijing ordered the deal reversed. The story broke late Saturday and is still developing.

Why it matters: Manus is a recognized general-purpose AI agent on the commercial market. If the acquisition falls apart, Manus's roadmap, funding, and competitive position are all up in the air — and the next buyer will face the same geopolitical scrutiny.

AI coding agents find the right file but miss the exact lines that matter, study shows

The Decoder · June 14

Research covering multiple agent architectures reveals a persistent granularity gap: agents locate the correct file but fail to pinpoint the right lines for modification.

Why it matters: This is a structural limitation, not a model-quality issue. Every coding agent user has experienced this — the agent "finds" the buggy area but edits the wrong line. Fixing this requires fundamental changes in how agents reason about code structure, not just bigger context windows.

Making Claude a Chemist

Hacker News · June 14

Anthropic demonstrates Claude autonomously planning and executing multi-step chemical synthesis workflows, from reagent selection to spectroscopy interpretation.

Why it matters: This is a template for domain-specific agentization of frontier models. If Claude can run a chemistry lab autonomously, the same pattern applies to biology, materials science, and drug discovery — turning LLMs into autonomous research assistants across hard sciences.

Amazon CEO's talks with U.S. officials triggered crackdown on Anthropic models

Hacker News · June 14

The Wall Street Journal reports that Amazon CEO Andy Jassy's conversations with US officials were the trigger for the government crackdown that forced Anthropic to disable Fable 5 and Mythos 5 globally.

Why it matters: This introduces a corporate-competitive angle to what had been framed as pure safety regulation. The largest investor in Anthropic may have been the one to set the regulatory machinery in motion, raising questions about conflicts of interest in AI governance.

OpenAI faces investigation from state attorneys general

TechCrunch AI · June 13

Multiple state attorneys general are jointly investigating OpenAI across advertising practices, health data handling, and consumer protection.

Why it matters: OpenAI is the operator of Codex, one of the most widely used AI coding agents, and GPT-5.5 powers countless agentic workflows. A multi-state investigation could lead to operational restrictions that ripple through every agent ecosystem that depends on OpenAI's models.