🌟 Vasilij’s Note
This week felt like a confirmation rather than a surprise. Jensen Huang named Claude Code as the inflection point that moved AI from answering to doing. Anthropic committed $100 million to getting consultancies into production. IBM spent $11 billion on the data problem that quietly kills most agent deployments before they even start. None of this is speculative anymore — the infrastructure is shipping, the partner programmes are live, and the forecast window Gartner is giving firms to act is three to six months. The pattern I keep seeing with clients is that the technology is no longer the bottleneck. The bottleneck is the decision to treat AI as infrastructure rather than a tool someone opens in a tab. That shift requires a different conversation internally — about ownership, about governance, about which workflow you start with.
In Today's Edition:
This Week in Agents | What Changed
NemoClaw ships at GTC 2026; Jensen Huang names Claude Code the agentic inflexion point. At the SAP Centre in San Jose on March 16, NVIDIA released NemoClaw, its enterprise-secure reference stack for OpenClaw agents with policy-based security, network guardrails, and a privacy router that decides where each inference executes. Huang framed the last two years in three chapters: ChatGPT, o1 reasoning, and then Claude Code as the moment AI moved from answering to doing. He said 100% of NVIDIA's engineers now use Claude Code, Codex, or Cursor. Adobe, Atlassian, Cisco, CrowdStrike, SAP, and Salesforce are all advancing agents on the NVIDIA Agent Toolkit.
Anthropic launches Claude Partner Network with $100M commitment — On 12 March, Anthropic launched a formal partner programme for consultancies and professional services firms deploying Claude at enterprise scale. The $100 million 2026 commitment funds training, sales enablement, co-marketing, and a fivefold expansion of Anthropic's partner-facing team. Accenture, Deloitte, Cognizant, and Infosys are anchor partners. A new technical certification — Claude Certified Architect, Foundations — is available immediately. A Code Modernisation starter kit targets legacy codebase migration, described as one of the highest-demand enterprise workloads.
Alibaba launches Wukong — multi-agent coordination for enterprise
On March 17, Alibaba released Wukong, an enterprise AI platform that coordinates multiple agents through a single interface. Currently in invitation-only testing, it handles document editing, approvals, meeting transcription, and research. It launches via DingTalk — Alibaba's workplace platform with over 20 million corporate users - with Slack, Microsoft Teams, and WeChat integrations planned. The launch sits inside a broader company reorganisation, with Wukong now under Alibaba's newly created AI Token Hub business group led by CEO Eddie Wu.
Top Moves - Signal → Impact
Anthropic vs the Pentagon — tech industry files amicus brief, court hearing set - Anthropic filed two federal lawsuits on March 9, challenging the DoD's supply chain risk designation. On March 12, the company sought an emergency stay from the DC Circuit Court, arguing the designation could cost it hundreds of millions to multiple billions in 2026 revenue. On March 13, major tech industry groups — representing hundreds of Pentagon contractors - filed an amicus brief calling for a pause, arguing the designation bypassed the procurement processes Congress created for exactly these situations. A court hearing is set for March 24.
→ This matters operationally for consultancies whose clients have DoD contracts, who will have been required to certify non-use of Claude. It also matters strategically: Anthropic launched the Partner Network the same week, signalling commercial momentum is not slowing down.
IBM completes $11B acquisition of Confluent — real-time data for agents — On March 17, IBM closed its $11 billion acquisition of Confluent, the Apache Kafka-based data streaming platform used by 6,500 enterprises, including 40% of the Fortune 500. IBM frames the deal as making real-time, governed data the engine of enterprise AI and agents. Day-one integrations span watsonx.data, IBM MQ, webMethods, and IBM Z. IDC estimates more than one billion new logical applications will emerge by 2028, all requiring live, continuously refreshed data.
→ Most agent deployments fail not because of model capability but because the data behind them arrives hours old and in silos. IBM is betting $11 billion that solving this is the unlock. That bet is almost certainly correct. IBM
Anthropic doubles Claude usage limits through March 27 — From March 13 to March 27, Anthropic is doubling usage limits during off-peak hours for Free, Pro, Max, and Team plans — automatically, no activation required. The promotion applies to Claude web, mobile, desktop, Cowork, Claude Code, and the Excel and PowerPoint add-ins. Enterprise plans are excluded.
→ Practically, this is a trial period by another name. IT leads can put delivery teams onto Claude across the full suite — including Cowork — gather real usage data, and make purchasing decisions based on observed behaviour rather than vendor demos. Worth using deliberately if your firm has not yet tested the Cowork workflows.
Upskilling Spotlight | Learn This Week
NVIDIA GTC 2026 sessions — agent infrastructure
Over 700 sessions are available on demand following this week's conference. The Monday keynote is the required context: Huang's framing of Claude Code as the agentic inflection point, the OpenClaw/NemoClaw enterprise security announcement, and the case for why "every company needs an agentic system strategy." Directly relevant for any consultancy advising on AI deployment architecture or building the business case internally. No paywall for session recordings.
Outcome: a concrete, technically grounded understanding of where the agent infrastructure stack stands right now, from the company that builds the hardware it runs on.
Anthropic Claude Partner Network — Academy materials and Code Modernisation starter kit
Now accessible via the Partner Portal upon joining the network (free). The Code Modernisation starter kit is the practical entry point: a structured approach to migrating legacy enterprise codebases using Claude's agentic coding capabilities. If your firm does any software delivery work, this is the highest-demand enterprise workload identified by Anthropic's own commercial team.
Outcome: a ready-to-use framework for scoping and delivering Claude Code engagements with enterprise clients.
Maker Note | What I built this week
This week, I ran the Cowork cross-app workflow - financial model in Excel, and structured a PowerPoint deck built from it in a single session.
Decision: it works precisely as described on clean, well-structured data, and breaks predictably on messy inputs, which is genuinely useful information to have before deploying it on a client engagement. Anthropic's own caveat stands: treat this as a directional signal and validate outputs before removing manual review.
Most consultancies are still running AI the old way - February 2026 Claude Cowork update changes that.
Operator’s Picks | Tools To Try
Claude Cowork (research preview)
Use for cross-app workflows between Excel, PowerPoint, and other Microsoft 365 applications in a single session. Context carries across applications without copy-paste. Available on all paid plans, Mac and Windows.
Standout: the plugin marketplace lets an admin create once and deploy across the whole team. Claude
NemoClaw (NVIDIA — now available) — Use for enterprise agent deployment with built-in policy-based security, privacy routing, and hardware-agnostic operation. Designed from the ground up for enterprise security posture, not retrofitted from a developer tool.
Standout: the privacy router decides whether inference runs on local GPU or cloud-based on rules you define — useful for firms handling client data under confidentiality obligations. Caveat: early release, evaluate stability before production use. NVIDIA
Deep Dive | Thesis & Playbook
Claude Cowork: What the February 2026 Update Actually Changes for Consultancies
Most consultancies are still running AI the same way they ran it in 2024. One person, one tab, one task at a time. The February 2026 Cowork release is the first credible answer to a question Anthropic's own head of Americas raised in a February enterprise briefing: if 2025 was meant to be the year AI agents transformed professional services work, and they did not, what actually changed?
On Paper
Claude now carries context between Excel and PowerPoint in a single session. You run a financial model analysis, ask for a presentation built from it, and the deck is built from that same context. No export. No copy-paste. One continuous workflow.
The private plugin marketplace lets admins bundle skills, connectors, slash commands, and sub-agents into one deployable package. Create once — the whole team gets it on day one. The organisation owns the plugin file. It is not locked to Anthropic's servers.
Wall Street Prep benchmarked Claude at 5.5 out of 6 for financial analysis tasks, against Copilot at 4.4 and ChatGPT at 2.5. This is where the capability gap shows up in actual delivery work.
In Practice
Cross-app workflows perform as described on clean, well-structured data. Messy spreadsheets produce messy outputs. That is not a criticism specific to Cowork — it is the same rule that applies to every automation. The difference is that the failure mode is now more visible, which is actually useful.
The plugin marketplace is only valuable if someone in your firm can build and maintain the plugins. Most consultancies do not have that person yet. Two paths: invest in the capability now and get ahead of the field, or wait six months for the ecosystem to mature and the gap to the early movers to widen.
Setup reality does not change because the product improved: 5–15 hours before value is delivered, 2–4 hours monthly maintenance, 2–4 weeks before consistent adoption across delivery teams.
Issues / Backlash
Anthropic has been explicit that cross-app workflows are a directional signal, not a finished capability. Validate against your existing workflows before removing manual review from client-facing outputs.
Live data connections — FactSet, MSCI, LSEG — are available but require separate setup through the connector directory. The promotion period running to March 27 is the right time to test this at no extra cost before committing.
The same governance questions that apply to any agent deployment apply here. Plugins that touch client data need approval workflows and audit logs before production use.
My Take (What to do)
Startup (15–40 staff):
Start with the Excel-to-PowerPoint workflow this week. The promotion period through March 27 gives you double the usage limits across the full Cowork suite. Run it against a real client brief, not a synthetic test. The results will tell you more than any demo.
SMB (50–120 staff):
Assign one person to own the plugin build — not a new role, make it explicit within an existing ops team member's responsibilities. Use the Partner Network's Claude Certified Architect certification to build the internal credibility for that person's decisions. Establish an approval workflow before deploying firm-wide.
Enterprise (150–250 staff):
You are competing with firms that have deployed Claude to 30,000–350,000 staff. Cognizant, Accenture, TELUS, and Infosys are already in the Partner Network with dedicated Claude business groups. Join the network this week — it is free and gives you access to the same playbooks those firms are using. Run a formal business case with a 12-month payback minimum. Governance framework before firm-wide deployment.
How to Try (15-minute path)
Download Claude Desktop and enable "Upgraded file creation and analysis" in experimental settings. (2 min)
Open a real financial model in Excel. Ask Claude to analyse it across the key tabs and identify the three most significant trends. (5 min)
In the same session, ask Claude to build a structured presentation from that analysis. Observe whether context carries from step 2. (5 min)
Note where it works and where it breaks. Document the gaps against your firm's actual workflow.
Success metric: a clear yes or no on whether the Excel-to-PowerPoint workflow is ready to replace manual steps in your delivery process — based on observed behaviour, not vendor claims.
Spotlight Tool | Claude Cowork
Purpose: cross-application AI workflows for Microsoft 365, designed for firm-wide deployment via private plugin marketplace.
Edge: context carries between Excel and PowerPoint in a single session. Admin creates once, the whole team gets it on day one. Organisation owns the plugin file — not locked to Anthropic's servers.
Cross-app workflow (Excel → PowerPoint)
Private plugin marketplace with admin controls
Firm-wide deployment without individual setup
Compatible with Claude Agent SDK
Available on all paid plans, Mac and Windows
Try it: claude.ai/download
What did you think of today's email?
Sponsored - Partner
n8n – An open‑source automation platform that lets you chain tools like DeepSeek, OpenAI, Gemini and your existing SaaS into real business workflows without paying per step. Ideal as the backbone for your first serious AI automations. Try: n8n
Did you find it useful? Or have questions? Please drop me a note. I respond to all emails. Simply reply to the newsletter or write to [email protected]