🌟Vasilij’s note
This week I recorded a complete guide to Claude Cowork Skills, Plugins, and Connectors - live, on camera, no script polish. Not a product walkthrough. My actual methodology for building reusable consulting ops tooling: the discovery-to-proposal workflow, the plugin bundle that wires six skills into a single deployment, the scheduling that removes humans from repetitive initiation. The reaction I always get when I show this: people clock the tools. The ones who get it clock the design pattern underneath - scope first, scaffold second, test on real inputs third, iterate over days not minutes. That pattern is the unlock. It transfers to any workflow, any firm size. Meanwhile, Microsoft and OpenAI ended their exclusivity arrangement, Coinbase and Freshworks publicly attributed layoffs to AI, and IBM found that 76% of companies now have a Chief AI Officer versus 26% last year. These are not product stories. They are structural signals. The firms that read them correctly will be advising clients on workflow redesign. The ones that don't will be competing with smaller teams doing the same work with agents.
In today's edition
This week in agents | What changed
Claude for Small Business launches
Anthropic released a dedicated Small Business plan on 13 May, bringing Claude's agentic capabilities within reach of firms that couldn't justify enterprise pricing → lowers the barrier for 15–40 person consultancies to access Skills, Connectors, and Cowork features without enterprise negotiation cycles. Today we are talking about Plugins and Skills, let me know if you would like me to cover this SME Plugin from Anthropic.
Microsoft and OpenAI end exclusivity
On 27 April, the two companies restructured to a non-exclusive licence model. OpenAI can now sell its models through any cloud provider, not just Azure. Microsoft retains a 27% stake and a licence through 2032, but the AGI clause is gone and revenue share is capped -> the AI infrastructure moat is collapsing. For consultancies choosing a model provider, lock-in risk fell materially. The correct question is now which model performs on your actual workflows, not which vendor partnership looks most stable.
Coinbase and Freshworks cut 1,200+ roles and blame AI directly
Coinbase CEO Brian Armstrong announced 700 redundancies on 5 May, stating AI lets smaller teams do in days what took weeks. The company is replacing traditional departments with "AI-native pods," including one-person teams directing agents across engineering, design, and product. Freshworks made similar cuts (11%, ~500 roles) the same week for the same stated reason -> the first firms publicly and explicitly restructuring around AI, not bolting it on. For consultancies: this is the client-side conversation that is arriving.
Top moves | Signal → impact
76% of companies now have a Chief AI Officer - and 93% say culture is the blocker, not technology
IBM's survey of more than 2,000 organisations found the share of firms with a dedicated CAIO jumped from 26% in 2025 to 76% in 2026. The same report found cultural resistance, not access to models or tooling, is the primary adoption barrier. -> For consultancies: the governance and change management gap is now the commercial opportunity. Firms that help clients redesign workflows and retrain working habits are more valuable than firms that recommend another tool. Most businesses do not need an AI recommendation. They need someone who can make their teams actually use it consistently.
Anthropic and Blackstone, Hellman & Friedman, Goldman Sachs build enterprise AI services company
Announced 4 May, the partnership creates a dedicated enterprise AI services entity combining Anthropic's models with heavyweight professional services distribution. -> For mid-market consultancies: the enterprise channel for Claude is professionalising fast. Firms that build deep delivery capability around Claude workflows now have a route to enterprise accounts via partners, not just direct sales.
OpenAI launches a $4 billion enterprise deployment company and acquires consulting firm Tomoro
Announced 11 May, the OpenAI Deployment Company (DeployCo) launched backed by more than $4 billion from 19 partners including TPG, Goldman Sachs, McKinsey, Bain & Company, and Capgemini. The model: embed Forward Deployed Engineers directly inside client organisations to move AI from pilot to production. The Tomoro acquisition brings 150 engineers with enterprise clients including Tesco, Virgin Atlantic, and Red Bull on day one. OpenAI's applications CEO told staff internally that Anthropic's enterprise gains should be a "wake-up call" and that the company cannot "miss this moment because we are distracted by side quests." -> The race to own enterprise AI deployment is now a $4 billion funded land grab by OpenAI, a $1.5 billion joint venture by Anthropic, and incumbents including McKinsey and Capgemini sitting on both sides of the table. For mid-market consultancies: the window to build specialised AI delivery capability before the major players commoditise implementation is narrowing. Differentiation now comes from depth of domain knowledge and speed of deployment, not access to the models.
Meta shifts from open source champion to proprietary frontier model
This week Meta announced $115-135 billion in AI capex for 2026 while quietly releasing a proprietary frontier model that outperforms parts of its own Llama 4 mid-size lineup at lower compute cost. The model performs strongly across multimodal reasoning, health tasks, and agentic workflows. Meta spent years positioning itself as the open source alternative to OpenAI - this move signals the competitive reality behind that strategy. -> For developers and consultancies building on Meta's open models: watch the proprietary roadmap closely. Open source was always partly strategic. If the centre of gravity shifts toward closed systems, teams relying on Llama for self-hosted deployments need a contingency. Evaluate your model dependency mix now, before a licensing or capability shift forces the conversation.
Upskilling spotlight | Learn this week
Claude for Financial Services: A Practical Deployment Guide (Anthropic)
Freshly published by Anthropic this week, this covers the full product stack for regulated industries: Claude Cowork for document-heavy project work, Claude Code for quantitative and engineering teams, Claude for Microsoft 365 for analysts in Excel and PowerPoint, and Claude Managed Agents for custom application builds. Includes customer examples, rollout sequencing, and governance requirements for firms operating under financial regulation. Even if you are not in financial services, the rollout sequencing and product routing logic (which Claude product for which workflow) applies directly to any professional services deployment.
Strategic/commercial focused Anthropic Claude Enterprise Licensing Guide 2026 (Redress Compliance)
A procurement and commercial strategy guide for firms advising clients on - or negotiating - Claude enterprise agreements. Covers the seat vs. API consumption split (API spend dominates the bill within 12-18 months of production deployment), model routing logic to reduce inference costs by 40-60%, and commit tier benchmarks ($250K-$1M unlocks 10-15% discount). Useful for any consultancy being asked by clients how to structure their AI vendor agreements, or managing their own Anthropic spend.
Maker note | What I built this week
This week I built and recorded the complete Skills, Plugins, and Connectors guide for Claude Cowork - a discovery-to-proposal skill from scratch, a six-skill Consulting Ops plugin, and a scheduled morning brief. The video covers every step of The Skill Playbook: Scope, Plan, Scaffold, Write, Test, Iterate.
Decision: Start with the /proposal skill before touching plugins. The ROI maths are immediate - 30 minutes to build, three hours saved per proposal, ten proposals a month. That is £4,500 in recovered partner time monthly at conservative rates. Prove the pattern on one skill before wiring six into a plugin bundle.
Operator’s picks | Tools to try
Claude Cowork - Skills tab
Use for building reusable firm-specific workflows: proposal assembly, client meeting summaries, onboarding packs, status updates. The SKILL.md structure (metadata, workflow, constraints) is the operative unit - one file, three sections, the rest is references and optional scripts.
Standout: the slash command (/proposal, /summary, /status) removes guesswork when multiple skills are loaded. Scheduling lets skills run without manual initiation.
Caveat: skills work on real inputs, not synthetic ones - test with actual messy discovery notes, not clean examples.
Tally Forms with conditional logic
Use for client intake automation without custom development. Pair with HubSpot for CRM sync - two minutes to configure.
Standout: free tier unlimited submissions; conditional logic handles complex intake flows without engineering.
Pair with: a /discovery skill that takes the Tally output and produces a structured summary ready for proposal.
Deep dive | Skills and Plugins: The Consulting Ops Stack
A 30-person consultancy typically starts each engagement week the same way: a partner opens five tabs, assembles context from three tools, writes a proposal from scratch, formats it to brand, sends it two days late. The prospect has spoken to a competitor. The problem isn't capacity - it's that every step is manual, inconsistent, and tied to whoever happens to have bandwidth.
Claude Cowork Skills and Plugins are the structural answer to this. Not a productivity feature. An operating model.
On paper
A Skill is a folder with a SKILL.md file containing three sections: metadata (the trigger description Claude reads to decide whether to load the skill), workflow (the step-by-step process), and constraints (guardrails - British English, professional tone, two-page limit, whatever your standards require). Optional references folder holds your brand guide, proposal templates, pricing frameworks. Optional scripts folder handles technical tasks like API calls or formatted document generation.
A Plugin is a bundle: multiple skills, slash commands, subagent configurations, hooks, and MCP connector configs. Install once, the whole team gets the same toolkit, the same connectors, the same process. Plugins are distributable - you can hand them to a client at the end of an engagement as a leave-behind, or package them as a paid product.
The Consulting Ops plugin described in this week's video bundles six skills - /discovery, /proposal, /onboard, /summary, /status, /case-study - connected to Gmail, HubSpot, Notion, and Google Calendar. One plugin. One install. The full client lifecycle.
In practice
The /proposal skill took 30 minutes to build end-to-end, including two iterations. Real test on messy discovery notes: structure consistent, detail proportional to input quality. The skill enforces formatting regardless of how rough the input is - that is the operational value, not the output quality on a clean prompt.
The Consulting Ops plugin took approximately two minutes to scaffold once the individual skills existed. The full workflow - discovery notes in, structured summary, full proposal, Notion pages created, follow-up email drafted in Gmail - ran in 90 seconds on a real test. Manually, the same sequence takes half a day of partner admin.
Scheduling compounds the value. A daily morning brief, a weekly pipeline summary, a Monday capacity check - these are things firms already do, manually, at low efficiency. Moving them to scheduled skills recovers 5-10 hours a week per senior operator.
Issues/backlash
Skills require real scoping work upfront. Firms that skip the Scope step and jump straight to building get generic outputs that don't justify the effort. The six-step Skill Playbook exists precisely because prompt-only skills without a documented process produce inconsistent results.
Plugin distribution assumes clean connector authentication. Per-user authentication means team-wide deployment requires individual setup - this does not scale cleanly above 30 staff without a rollout plan. The quality of the prompt templates attached to each skill tile matters as much as the infrastructure. A poorly written template produces generic output that does not justify the click.
Security posture carries over from the previous edition's warnings: prompt injection, data access permissions, and shadow skill sprawl are live risks at production scale. Treat skills containing client data like any other third-party integration - security review before rollout, not after an incident.
My take (what to do)
Startup (15-40 staff): Build one skill this week. Not a plugin. One skill. Pick the workflow that happens most frequently and costs the most partner time - for most firms that is proposal assembly. Follow the Skill Playbook: Scope what you actually do manually, step by step. Plan what references you need (your best past proposal, your brand guide, your pricing framework). Scaffold with Claude, write the SKILL.md, test on three real inputs. Calculate ROI before building anything else: (time saved per proposal x number per month x 52 weeks x partner hourly rate) - setup time. If it is positive within six months, proceed to the next skill. If not, the workflow is not ready.
SMB (50-120 staff): You need a skills owner - one delivery lead responsible for the portfolio. Not an additional role; make it explicit within an existing ops team member's remit. Their first four weeks: build the /proposal and /summary skills, document what works and what breaks on real inputs, replicate for two colleagues before any wider rollout. Track revision rounds per output. If the skill produces a usable draft without revision on fewer than six in ten real inputs, the SKILL.md needs another iteration pass. Only move to plugins once individual skills are stable - plugin complexity before skill stability creates maintenance debt.
Enterprise (150-250 staff): The governance question precedes the build question. Map which data sources will connect to which skills. Classify those sources. Verify your existing AI usage policy covers skill deployments. Establish a central skills registry - a Notion page is sufficient - logging which skills are deployed, who approved them, and what data they access. Require security review for any skill touching client or commercial data. Rollout to full team only after per-user connector authentication is resolved cleanly at pilot scale.
How to try (15-minute path)
Download the Consulting Ops plugin from the video description. Open Claude Cowork, navigate to Plugins, upload the file. (3 min)
Connect HubSpot, Notion, and Gmail in Connector settings. These three cover the majority of the Consulting Ops plugin's data sources. (5 min)
Open a new task, select the Consulting Ops plugin, type /proposal, paste the messiest set of discovery notes you have. Observe the output - structure, tone, pricing section. Note how many revisions it needs before you would send it. (7 min)
Success metric: a proposal draft that requires fewer revisions than your current manual process. If it produces a usable first draft in under two minutes, the pattern is working. If it requires more correction than writing from scratch, the references folder needs your actual templates, not placeholders.
"AI is becoming capable of doing increasingly meaningful work inside organisations. The challenge now is helping companies integrate these systems into the infrastructure and workflows that power their businesses."
Spotlight tool | Claude Cowork - Skills and Plugins
Reusable, portable workflow packages that turn your firm's best processes into deployable infrastructure - not one-shot outputs buried in chat history.
-> Skills: SKILL.md structure with metadata, workflow, constraints; references folder for brand assets; optional scripts for technical tasks
-> Plugins: bundle multiple skills with slash commands, connectors, and subagent configs into a single installable package
-> Scheduling: run skills on a defined cadence without manual initiation - daily briefs, weekly summaries, Monday pipeline checks
-> Distribution: export and share plugins across your team, hand off to clients as leave-behinds, or package as paid products
-> Slash commands: /proposal, /discovery, /summary, /status, /case-study - explicit triggers that remove guesswork when multiple skills are loaded
What did you think of today's issue?
Did you find it useful? Or have questions? Please drop me a note., I respond to all emails. Simply reply to the newsletter or email [email protected].
This issue’s sponsor
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.
Refer and win
Share this newsletter for a chance to win!