This Week in AI & Automation

Week of March 29, 2026

Last week was about chips and agents. This week is about the layer underneath: data. IBM closed its $11 billion Confluent acquisition, betting that real-time data streaming is the missing piece for production AI deployments. Microsoft shipped agentic business applications in Power Platform that let users express intent while AI agents handle execution. McKinsey published new numbers showing 12% of job tasks are already automated by AI — but 8% of new job categories were created in the same period. Nexthop AI raised $500 million for the networking fabric connecting GPU clusters. And the Colorado AI Act set a June 30 compliance deadline that will force enterprises to audit algorithmic decision-making. The AI automation news this week points to one theme: the infrastructure layer beneath the models is where the real value accrues.

The Big Story

IBM Closes $11B Confluent Deal — Real-Time Data Becomes the AI Bottleneck

IBM completed its acquisition of Confluent on March 17 for $31 per share, an enterprise value of roughly $11 billion. Confluent's data streaming platform powers real-time operations at over 6,500 enterprises, including 40% of the Fortune 500. IBM is integrating it with watsonx.data, MQ, webMethods, and IBM Z to build what it calls a "smart data platform" for AI agents and automated workflows.

The thesis is straightforward: as enterprises move from AI experimentation to production, the critical barrier is data — clean, governed, continuously refreshed, and delivered at the speed AI demands. In most enterprises today, data is still siloed across systems, arriving hours or days after it was generated. AI models running on stale data make stale decisions.

Confluent streams live operational events directly into watsonx.data, ensuring every model, agent, and workflow runs on continuously updated enterprise data with lineage, policy enforcement, and quality controls built in. The Hart-Scott-Rodino waiting period expired without a second request from antitrust regulators, signaling that even Washington sees this as a market-consolidation play, not a competition concern.

Source: IBM Newsroom, Yahoo Finance

Our Take: This acquisition validates what we see in every enterprise AI project: the model is 20% of the work, and the data pipeline is 80%. IBM is making a massive bet that whoever owns the real-time data layer wins the AI agent era. For companies building AI agents that need to act on current information — processing invoices, detecting fraud, managing inventory — real-time streaming is not optional. It is the difference between an AI that automates and an AI that just summarizes yesterday's reports.

Notable Developments

Microsoft Ships Agentic Apps in Power Platform — Intent Replaces Workflow

Microsoft's March 2026 Power Platform update introduced agentic business applications: autonomous systems that can make decisions, learn from interactions, and execute complex business processes without constant human oversight. The update includes enhanced governance controls, AI-powered development tools, and new admin controls for agent security.

The philosophical shift is significant. Instead of teaching people how to use systems, organizations now let people express intent — and let agents determine how that intent is carried out. Early implementations handle dynamic pricing adjustments, intelligent resource allocation, and predictive maintenance scheduling. The 2026 Wave 1 general availability rollout begins in April.

Source: Microsoft Power Platform Blog, Windows News

Our Take: Microsoft is embedding agentic AI into the platform that millions of enterprises already use. The "intent-first" model eliminates the traditional bottleneck where business users must learn complex software to get work done. For companies evaluating AI vs RPA, this blurs the line entirely — Power Platform agents can now handle the unstructured, judgment-heavy tasks that RPA could never touch.

McKinsey: 12% of Job Tasks Already Automated — But 8% of New Job Categories Are AI-Created

McKinsey's Global Institute published new findings showing that AI-powered agents and robots could generate roughly $2.9 trillion in annual U.S. economic value by 2030. The more immediate data point: 12% of current job tasks across the economy have already been automated by AI over the past two years, while 8% of new job categories created in the same period were directly AI-related.

Current technology could theoretically automate over half of U.S. work hours, but McKinsey's midpoint scenario projects one-quarter to one-third of work hours actually shifting. The report emphasizes that the future involves partnerships between people, AI agents, and robots — not simple replacement. Demand for AI fluency has grown sevenfold since 2023 and now appears in postings for roles employing roughly 7 million U.S. workers.

Source: McKinsey, Fortune

Our Take: The 12% automation figure aligns with what we see in client deployments. The functions getting automated first — detail orientation, quality assurance, inventory management, invoicing, SQL programming — are exactly the tasks where AI delivers the clearest ROI. The 8% new-category creation is the underreported side of the story. Companies that redeploy freed capacity into AI-augmented roles are pulling ahead. Those that only see AI as a headcount reduction tool are leaving the bigger value on the table.

Nexthop AI Raises $500M for the Networking Layer Beneath GPU Clusters

Nexthop AI closed an oversubscribed $500 million Series B on March 10, led by Lightspeed Venture Partners with Andreessen Horowitz joining, pushing the company's valuation to $4.2 billion. Nexthop builds high-performance Ethernet switches designed natively for AI workloads, delivering both off-the-shelf and custom switching solutions built on open-source operating systems like SONiC and FBOSS.

The raise signals that the networking layer connecting GPU clusters is becoming a standalone investment category. With hyperscalers committing over $650 billion in 2026 capex for AI infrastructure, the bottleneck is shifting from compute to connectivity. a16z called it "one of the largest infrastructure market opportunities in a generation."

Source: BusinessWire, SiliconANGLE

Quick Hits

• Colorado AI Act sets June 30, 2026 deadline: SB 24-205 requires companies deploying high-risk AI in hiring, lending, and healthcare decisions to conduct regular bias audits. Enterprises with algorithmic decision-making have three months to comply. Drata
• 67% of Fortune 500 now have at least one AI agent in production: Up from 34% in 2025, with customer service as the top use case at 42% of deployments. Crescendo AI
• Alibaba launches Wukong AI platform: The enterprise automation platform targets corporate users with intelligent workplace automation, pushing deeper into the enterprise AI market as Alibaba reports earnings. CryptoNomist
• Lio raises $30M for AI procurement agents: Series A led by Andreessen Horowitz. Lio's agents read documents, evaluate suppliers, negotiate terms, and complete transactions — processes that once took weeks now completed in minutes. TechCrunch

Numbers of the Week

What We're Watching

The data infrastructure layer is consolidating around AI use cases. IBM buying Confluent for $11B, Databricks partnering with Accenture (25,000 trained professionals), and Snowflake's recent AI features all point the same direction: whoever controls the real-time data pipeline controls the AI agent stack. For companies running production AI, the choice of data platform is becoming as strategic as the choice of model provider. Lock-in risk is real — evaluate carefully before committing.

State-level AI regulation is filling the federal vacuum. With no comprehensive federal AI law, states like Colorado, California, and New York are setting their own rules. Colorado's June 30 deadline for bias audits in high-risk AI will catch many enterprises off guard. Companies that build AI governance into their systems now will spend weeks adapting to new rules. Those that don't will spend months — or face enforcement actions.

The agent paradigm is going mainstream through existing platforms. Microsoft embedding agents in Power Platform, Salesforce shipping Agentforce, and ServiceNow launching AI agents all share the same playbook: don't ask enterprises to adopt new tools — inject agentic capabilities into tools they already use. This dramatically lowers the barrier to entry for enterprise AI adoption but also means vendor lock-in compounds. The platform you already use will be the first to offer you an AI agent — and the hardest to leave once you accept.

The Bottom Line

This week's theme is plumbing. IBM spent $11 billion on a data streaming platform. Nexthop raised $500 million for network switches. Microsoft rearchitected Power Platform around agent intent. None of these are glamorous headlines, but they are the infrastructure decisions that determine whether AI agents actually work in production or just work in demos.

The model wars get the attention. The data layer wars determine the outcome. Every enterprise AI deployment we have seen succeed had one thing in common: the team spent more time on data strategy than model selection. IBM is betting $11 billion that this pattern holds at scale. Based on what we see in the field, they are right.

Get This Week in AI & Automation delivered every Saturday.

Previous Editions

• This Week in AI & Automation: The Agent Infrastructure Week (Mar 28, 2026)
• This Week in AI & Automation: The Great Rebuild (Mar 21, 2026)
• This Week in AI & Automation: The Safety Standoff (Mar 14, 2026)
• This Week in AI & Automation: The Computer Use Era (Mar 7, 2026)
• This Week in AI & Automation: The 40% Cut (Feb 28, 2026)
• This Week in AI & Automation: The $100 Billion Week (Feb 21, 2026)
• This Week in AI & Automation: The Age of Agent Teams (Feb 14, 2026)

Have a story we should cover? Contact us.