Meta Buys Manus: Enterprise AI Agents Spark $1B Land Grab

So, What Actually Happened?

We scanned 190,000 articles this week, and the word that kept showing up in places it never used to was ”agent.” Not the human kind. The software kind. And not just in research labs anymore.

Meta paid up for Manus, a Chinese-founded AI startup building agentic AI infrastructure. Salesforce reported a record-breaking quarter driven by what they're calling the ”agentic pivot,” with 180 organizations already adopting Agentforce for IT service management. Meanwhile, a growing governance gap is opening up as enterprises rush these agents into core operations without the frameworks to manage them. And behind all the software excitement, a Plug and Play analysis argues that AI's real next leap depends not on code, but on breakthroughs in silicon photonics and wide-bandgap materials.

The Bottom Line: Everyone's shipping agents. Almost nobody's governing them. And the hardware they'll eventually need to run on doesn't exist yet. That's the gap between ambition and infrastructure, playing out in real time.

The Tracks That Matter

1. Meta Buys Manus: The Agentic AI Acquisition Nobody Expected

The agentic AI market just got its clearest price signal. Meta acquired Manus, a Chinese-founded startup that built its reputation on autonomous AI agent infrastructure, the kind that can plan, execute, and iterate on complex tasks without human hand-holding. The deal underscores a shift: Big Tech is now buying agentic AI capabilities rather than building them from scratch.

The timing matters more than the price tag. This acquisition comes during a week when enterprise AI spending hit record highs and Salesforce reported its best quarter ever, driven almost entirely by AI agent products. Manus wasn't a research project. It was a functioning agent platform that enterprises could deploy. That's what Meta is buying: the ability to ship, not the ability to research.

The competitive angle is straightforward. Every major platform company now needs an agent story. Salesforce has Agentforce. The hyperscalers are building their own orchestration layers. Meta, by acquiring Manus, is signaling that it wants to play in enterprise agent infrastructure, not just consumer social AI. The Chinese founding is notable too: talent that cut its teeth building for one of the world's most competitive AI markets is now inside a Western Big Tech company.

What makes this more than a talent acquisition: Manus had paying customers. It had a platform. It had integration patterns that enterprises were already using. This isn't acqui-hiring. This is buying a working product and the team behind it.

Here's what works: If you're building on or evaluating AI agent platforms, this acquisition changes the competitive map. Meta's entry into enterprise agents means more options, but also more platform risk. Evaluate whether your current agent infrastructure is portable, or whether you're locked into a stack that just got a new competitor.

2. Salesforce's Agentic Pivot: 180 Organizations and a Record Quarter

The numbers tell one story: Salesforce posted a record-breaking 2026 performance, and the engine behind it is a product that barely existed 18 months ago. The deeper story is what they're calling the ”agentic pivot,” a strategic bet that AI agents deployed directly into business workflows, rather than chatbots answering questions, would drive the next wave of enterprise software growth.

The proof point landed this week: 180 organizations have now adopted Agentforce for IT service management. ITSM is a revealing first target. It's repetitive, expensive, and high-volume. The kind of work that agents can handle autonomously while freeing human teams for exceptions and escalations. Salesforce rising 87% in our influence tracking isn't hype; it reflects real enterprise adoption at scale.

The broader implication: Salesforce is proving that agents sell when they're embedded into existing workflows rather than positioned as standalone products. They didn't build a separate AI tool. They made the CRM do things autonomously. That's a different go-to-market than ”here's our AI assistant.”

Accenture shares jumped on a new AI partnership announcement the same week, signaling that the services ecosystem around enterprise agents is accelerating too. When the consultancies are hiring for it and the platforms are shipping it, that's not a trend. That's a market.

Here's what works: If your ITSM operation still runs on tickets-and-humans, you're leaving money on the table. Evaluate Agentforce and its competitors not as AI experiments, but as operational cost reduction tools. The 180 organizations already using it have a head start on the cost curve.

3. The Hard Reset: Why AI's Next Era Depends on Physics, Not Software

While everyone debates model architectures and training techniques, a Plug and Play analysis argues that AI's real bottleneck has shifted. The next era of AI depends on physical breakthroughs in semiconductors, not better algorithms. Silicon photonics, wide-bandgap materials like gallium nitride and silicon carbide, and novel memory architectures are the foundations that need to move before AI can scale further.

The companies positioning for this shift are telling. Onsemi is investing in wide-bandgap semiconductors for power efficiency. Aura Semiconductor is working on connectivity chips designed for AI edge devices. RAAAM Memory is attacking the memory wall with new architectures. And Syenta is building the system integration layer that connects these components. These aren't household names, and that's the point. The next AI infrastructure layer is being built by companies most people haven't heard of.

ASML's strategic pivot toward AI reinforces the same thesis. The company that makes the machines that make the chips is now explicitly orienting its roadmap around AI demand. When the toolmaker pivots, it's because the customers have already committed.

Data Pipelines, our top foundational technology this week (appearing across 59 articles with 16% growth in structural importance), underscores the pattern. The infrastructure layer everyone depends on is the one nobody headlines. Software AI gets the press. Physical AI gets the investment.

Here's what works: If your AI strategy assumes compute availability will keep scaling linearly, stress-test that assumption. Look at who's investing in semiconductor materials, memory architectures, and edge silicon. The next bottleneck won't be algorithmic. It will be physical, and the companies solving it now will set the terms.

4. Why Pure Data Mesh Breaks at Enterprise Scale (And What Actually Works)

Here's a contrarian take that earned its spot: HackerNoon published an analysis arguing that pure data mesh, the beloved decentralized architecture pattern that promised to liberate data teams from centralized bottlenecks, breaks down when you scale it across a real enterprise. Not because the theory is wrong, but because the organizational reality doesn't match the architectural ideal.

The argument is practical, not theoretical. Data mesh assumes domain teams will take ownership of their data products. In practice, most domain teams don't have the skills, the tooling, or the incentive structures to maintain production-grade data pipelines. The result: data quality degrades, discoverability suffers, and the ”mesh” becomes a collection of loosely connected silos that are harder to govern than the monolith they replaced.

What works instead, according to the analysis, is a hybrid approach: centralized governance and platform infrastructure with decentralized domain ownership. Think of it like a record label. The label handles distribution, quality standards, and marketing (centralized platform). The artists handle the music (domain ownership). Neither works without the other.

This connects to a broader pattern in our data. Data Integration ranks #2 in foundational importance this week across 63 articles, with 14% growth. Data Management ranks #4 across 59 articles. The infrastructure that actually moves data between systems, regardless of architectural pattern, is where the real work happens.

Here's what works: If you've been sold on pure data mesh, audit your domain teams' actual capabilities. Do they have dedicated data engineers? Do they have SLAs for data quality? If the answer to either is ”not really,” your mesh is probably a mess. The hybrid model (centralized platform, decentralized ownership) matches how organizations actually work.

5. The Governance Gap: Enterprises Ship Agents Faster Than They Can Manage Them

The governance gap is widening. A new analysis with Gartner input documents what anyone paying attention has suspected: enterprises are deploying AI agents into core business operations, from customer service to supply chain, far faster than they're building the governance frameworks to manage them. The gap isn't theoretical. It's operational.

The problem is structural. When an AI agent autonomously processes a customer refund, modifies a supply chain order, or escalates a security incident, who owns the decision? Traditional governance models assume a human in the loop. Agents break that assumption. ExchangeWire's analysis of automation vs. control in the advertising industry documents the same tension: WPP just restructured and cut staff as agencies shift to AI-driven models, but the governance for those AI-driven decisions is still being written.

The timing makes this worse. This governance gap is opening exactly as Salesforce ships Agentforce to 180 organizations, Meta acquires an agent platform, and cross-device agents approach practical reality. The deployment curve is exponential. The governance curve is linear.

The risk isn't that agents will do something catastrophic. It's that they'll do something wrong at scale, consistently, before anyone notices. A human makes a mistake and it affects one transaction. An agent makes a mistake and it affects thousands before the error surface becomes visible.

Here's what works: Before deploying any AI agent into a production workflow, define three things: decision authority (what can the agent decide alone?), escalation triggers (what always goes to a human?), and audit trails (how do you reconstruct why the agent did what it did?). If you can't answer all three, you're not ready to deploy.

6. Datadog and Sakana AI: When Observability Meets AI Research

A partnership that flew under most radars this week deserves attention. Datadog and Sakana AI announced a strategic partnership to advance AI innovation and observability for enterprises. Sakana AI, a Japanese AI research lab that appeared in our knowledge graph for the first time this week, is the interesting half of this equation.

Sakana AI entered our tracking as a new entity with immediate partnerships, the signature of a company that's been building quietly and is now stepping into the light. Datadog's interest makes strategic sense: as enterprises deploy more AI agents and models into production, the observability layer needs to understand AI-native workloads, not just traditional infrastructure metrics. CPU utilization doesn't tell you if your agent is hallucinating.

The broader pattern: observability is becoming the governance layer for AI in production. When you can't manually review every agent decision, you need monitoring that understands AI-specific failure modes, model drift, confidence degradation, and unexpected behavioral patterns. Datadog is betting that AI observability is the next infrastructure category, and Sakana AI brings research depth that a monitoring company doesn't have in-house.

Here's what works: If you're running AI workloads in production, evaluate whether your current observability stack understands AI-specific metrics. Traditional APM tools monitor latency and errors. AI observability needs to monitor model behavior, output quality, and decision patterns. The tools are catching up to the need, and this partnership signals where the market is heading.

7. FDA Quietly Makes AI in Drug Manufacturing Official

While the tech industry debates AI governance in the abstract, one regulator just made it concrete. The FDA's Center for Drug Evaluation and Research (CDER) added AI/ML guidance to its regulatory agenda, including new guidance on digital technologies and AI applications in pharmaceutical manufacturing. This isn't a position paper. It's a regulatory agenda item that will shape how drugs are made.

The significance is in the specificity. The FDA isn't talking about AI in general. They're targeting pharmaceutical manufacturing: quality control, process optimization, and production monitoring. This is the kind of regulatory clarity that turns AI from a pilot project into a production requirement. When the FDA says ”we're writing guidance for AI in manufacturing,” it means pharmaceutical companies need to start building compliance-ready AI systems now, not after the guidance drops.

Venture capital investment in AI-driven pharmaceutical applications is already accelerating, and a separate analysis documents why AI adoption in healthcare has been slow despite the investment. The FDA guidance could be the forcing function: when the regulator provides a framework, companies that have been waiting for clarity can finally move.

This connects to a broader pattern. Data Security ranks #3 in foundational importance this week across 60 articles. In pharma, data security isn't an IT problem; it's a regulatory requirement. AI systems in drug manufacturing will need to demonstrate data integrity, audit trails, and validated outputs. The companies that build these capabilities early will have a compliance moat.

Here's what works: If you're in pharma or adjacent regulated industries, start mapping your AI use cases against the emerging FDA framework now. Don't wait for final guidance. Build your validation infrastructure, your audit trails, and your quality systems today. The guidance will reward companies that are already doing it right.

Signal vs. Noise

🟢 Signal: Sakana AI appeared in our knowledge graph for the first time this week, and immediately registered partnerships with Datadog. When a specialized AI research lab emerges from stealth directly into enterprise partnerships, that's not hype. That's a company that spent its quiet phase building something companies want to buy. The pattern of Japanese AI research labs forming Western enterprise partnerships is worth tracking; it suggests a geographic diversification of AI innovation that goes beyond the US-China axis.

🔴 Noise: Enterprise AI spending ”records” keep getting announced, but governance frameworks aren't keeping pace. This week, OpenAI's $40B round generated headlines about AI spending reaching new heights. But spending isn't progress. When the governance gap analysis shows enterprises deploying agents into core operations without decision authority frameworks, escalation triggers, or audit trails, the ”record spending” narrative starts looking like the housing market in 2007: everyone's buying, nobody's inspecting.

From the 190K

We scanned 190,000 articles this week. Here's what no one's talking about:

The Agentic Convergence Is Happening Across Every Industry Simultaneously

Five separate stories this week, from five unrelated industries, shared one common thread: AI agents are moving from demos to deployment, everywhere, at the same time. Salesforce shipped Agentforce to 180 ITSM organizations. Meta acquired an agent infrastructure company. Researchers published GUI-Owl-1.5, bringing cross-device AI agents closer to practical reality. eInfochips launched NomAIzo for edge-to-enterprise AI adoption. And RobosizeME raised $2M to bring AI workflow automation to the global hotel industry.

Individually, these are footnotes. Together, they reveal a phase transition. Agents aren't an enterprise software category anymore. They're an operating model. IT service management, consumer platforms, scientific research, edge computing, and hospitality all moved toward agent-driven workflows in the same 48-hour window. When every industry pivots to the same pattern simultaneously, that's not a trend. That's a platform shift. The last time something moved this fast across industries was cloud adoption in 2012-2014, and the companies that waited until 2016 spent years catching up.

🔍 Below the surface: Data Analytics appeared in 79 articles this week but made zero mainstream headlines. Here's how you spot real infrastructure: when something shows up everywhere but headlines nowhere, it means every company is doing it and nobody thinks it's interesting enough to write about. Data Analytics has the highest foundational importance of any entity in our knowledge graph this week, growing 15% in structural importance. The boring layer everyone depends on, nobody talks about.

By The Numbers

• $40B — OpenAI's latest funding round as enterprise AI spending hits record highs
• 180 organizations — Enterprises now running Agentforce for IT service management, up from near zero 18 months ago
• 87% influence growth — Salesforce's PageRank surge, reflecting real enterprise adoption not just mentions
• 79 articles on Data Analytics — Highest foundational importance in our knowledge graph, zero mainstream headlines
• +16% Data Pipelines growth — Largest structural importance gain among foundational technologies
• $2M seed for RobosizeME — AI workflow automation reaches the global hotel industry
• 57 articles on AI — Artificial Intelligence concept grew 93% in influence this week, driven by agent adoption stories
• 60 articles on Data Security — Third-highest foundational importance, 11% growth, still no mainstream headlines

Deep Dive: The Agentic Enterprise Is Here. The Guardrails Aren't.

Like a DJ watching the entire dancefloor change rhythm at once, this week felt different. Not because one company made a move. Because every company made the same move. Agents went from ”interesting experiment” to ”operating model” in about 72 hours, and the infrastructure to manage them is still being drawn on whiteboards.

The Deployment Tsunami

Here's what happened in a single week: Salesforce shipped Agentforce to 180 organizations for ITSM. Meta bought an entire agentic AI company. A research team published cross-device agents that can operate across phones, desktops, and tablets simultaneously. And an edge computing company launched a product to bridge agent intelligence from edge devices to enterprise systems. Every one of these stories represents agents doing real work, not generating demos.

The Governance Vacuum

And here's what didn't happen: nobody shipped governance. The analysis with Gartner input lays it bare. Enterprises are moving agents into core operations, customer service, IT management, supply chain, without decision authority frameworks, without escalation protocols, and without audit trails that can reconstruct why an agent did what it did. The advertising industry is already feeling the consequences: WPP restructured and cut staff as agencies shift to AI-driven models, but the governance for those autonomous decisions is still being written.

The Infrastructure Gap Below

Meanwhile, the physical infrastructure these agents will eventually need doesn't exist yet. Today's agents run on cloud inference. Tomorrow's agents, the ones making real-time decisions on factory floors and in autonomous vehicles, will need silicon photonics, new memory architectures, and edge chips that haven't been manufactured at scale. The software is outrunning the hardware by about three years.

What Actually Works

1. Define decision boundaries before deployment: Every agent needs a clear scope. What can it decide alone? What requires a human? Document this before the agent touches production data.
2. Build audit trails from day one: If you can't reconstruct why an agent made a specific decision, you can't debug it, defend it, or improve it. Logging isn't optional.
3. Monitor agent behavior, not just agent uptime: Traditional monitoring tells you if the agent is running. AI observability tells you if the agent is making good decisions. Those are different tools.
4. Stress-test your hardware assumptions: If your agent strategy assumes unlimited cloud inference, model what happens when latency matters or costs triple. The physical AI infrastructure gap is real.

The agentic enterprise is happening whether your governance framework is ready or not. The question is whether you're building the guardrails while the train is moving, or hoping the track doesn't curve until you're done. History says: it always curves.

What's Coming

Deloitte's State of AI 2026: The Enterprise Reality Check

Deloitte published its 2026 State of AI in the Enterprise report, providing one of the most comprehensive institutional views on where enterprises actually are with AI adoption (not where vendors say they are). When Deloitte surveys enterprise leaders, the gap between AI ambition and AI reality becomes quantifiable. Watch for the data on ROI realization rates, as that number tends to be more sobering than the spending headlines suggest.

Andreessen Horowitz Weighs In on AI Regulation

CSIS is hosting Andreessen Horowitz's Jai Ramaswamy and Matt Perault for a conversation on AI regulation and innovation. When the most aggressive AI investor takes a public position on regulation, it shapes the Overton window for everyone else. Given the Anthropic/Pentagon saga from last week and the governance gap documented this week, the timing is pointed.

Broadcom's Private Cloud Predictions: Sovereignty Returns

Broadcom published 2026 private cloud predictions focused on cost, sovereignty, and what they call ”the new application stack.” After last week's data sovereignty coverage, this is worth watching for signals about whether enterprises are actually pulling workloads back from public cloud or just talking about it. The sovereignty angle is especially relevant for organizations running AI on data subject to residency requirements.

For Your Team

Monday's meeting prompt: ”If every department deployed AI agents into their core workflows next quarter, and something went wrong at scale, could we reconstruct what happened and why? If the honest answer is no, what are we missing?”

The Agent Readiness Framework:

1. Decision authority mapping — For every process you're considering for agent automation, document what decisions an agent can make independently, what requires human approval, and what's completely off-limits. No ambiguity.
2. Audit trail architecture — Build logging that captures not just what the agent did, but why it did it: which inputs it considered, which rules it applied, and what alternatives it rejected. This is your compliance insurance.
3. Failure radius calculation — When an agent makes a mistake, how many transactions does it affect before someone notices? If the answer is ”thousands,” you need circuit breakers and anomaly detection before you deploy.
4. Hardware dependency stress test — Map your agent workloads against their infrastructure requirements. What happens if inference costs triple? What happens if latency requirements tighten? If you can't answer, you're building on assumptions.

Share-worthy stat: 180 organizations adopted Salesforce Agentforce for ITSM, up from near zero 18 months ago. That adoption curve is steeper than cloud computing's early trajectory, and unlike cloud, agents make decisions autonomously. The governance frameworks need to move just as fast.

Go deeper: Track AI agent adoption and governance trends in real-time →

The Track of the Day

”The gap between governance talk and governance action continues to widen, and most of the content is recycled position papers rather than new regulatory movement.”
— From our analysis of 190,000 articles this week

Everyone's got a governance framework. Nobody's got governance. That's the difference between a playlist and a performance. One looks good on paper. The other actually moves the room.

We scanned 190,000 articles this week so you don't have to. Data Pains → Business Gains.

Published: February 28, 2026 | Curated by Yves Mulkers @ Ins7ghts