🌟 Vasilij’s Note
This week reinforced something I see over and over with clients: agents aren't failing because the tech isn’t ready. They’re failing because we keep automating the wrong things, too early, without doing the boring maths. MCP going open is a big structural shift, but it doesn’t magically fix ROI, governance, or adoption. I keep thinking of a pattern: knowledge => workflows (processes) => agents. There is no shortcut to this. Subscribe, I will expand on this in the next few newsletters.
In Today's Edition:
This Week in Agents | What Changed
Anthropic donates MCP to Linux Foundation under new Agentic AI Foundation with backing from OpenAI, Block, Microsoft, Google, and AWS. Over 60,000 projects adopted the standard since August. → Open standard for agent integration reduces vendor lock-in risk for consultancies building workflows into client-facing systems. Anthropic
Q4 research from McKinsey, Deloitte, and Google Cloud reveals the pilot trap. 74% of executives achieve ROI within the first year, but only 10% of use cases scale past proof-of-concept. → The 64-point gap isn't technology—it's governance, process redesign, and adoption discipline. Google
Microsoft launches vertical agents for mid-market through Dynamics 365, targeting professional services with out-of-box workflows for sales orders and payables. Fujitsu case study claims 67% productivity gains. → Validation that agent-powered processes work at scale, and direct competition for consultancies selling automation services. Microsoft
Top Moves - Signal → Impact
MCP becomes open standard under Linux Foundation governance
Anthropic, OpenAI, Block co-found Agentic AI Foundation with support from Microsoft, Google, AWS, Bloomberg. Over 60,000 projects adopted MCP since August.
→ Standard way to connect agents to business systems means less vendor lock-in when integrating with your existing stack (HubSpot, Asana, SharePoint). Pre-built connectors reduce custom integration costs that kill ROI. ERP Today Linux
Security researchers flag MCP vulnerabilities
Prompt injection, tool permissions, and lookalike tool issues were discovered in MCP implementations during security audits.
→ Don't wait for incidents to build governance. Agent sprawl without guardrails creates audit nightmares and compliance exposure. Approval workflows and rollback mechanisms need to be designed now, whilst pilots are small. Bitsight
The insurance sector accelerates from 8% to 34% AI adoption in 12 months
325% increase in the regulated industry with strict compliance requirements. Automated underwriting, claims triage, and fraud detection workflows are driving adoption.
→ They didn't just automate tasks—they redesigned processes around agents. Document-heavy, repeatable workflows are where ROI shows up fastest for consultancies. Fenwick
Upskilling Spotlight | Learn This Week
Google Cloud's ROI of AI 2025 Report
Implementation patterns by industry, 74% ROI within first year statistics with methodology breakdown. Google
McKinsey's Agentic AI Advantage Report
Understand why horizontal tools (Copilot) spread value thinly vs vertical use cases (proposal generation) that impact bottom line. Mckinsey
Maker Note | What I built this week
This week I mapped three workflows in a client's operations—proposal assembly, client reporting, delivery handoffs. Timed each across two weeks of actual usage.
Decision: only one passed the 6-month ROI threshold when factoring in setup hours and maintenance. The others will wait until utilisation pressure justifies the investment.
I spent two years building workshops around ChatGPT. Last month, I stopped using it. Here's what changed when managing multiple client engagements simultaneously…
Operator’s Picks | Tools To Try
MCP servers for common platforms
Pre-built connectors for Google Drive, Slack, HubSpot, Asana eliminate custom integration work.
Standout: maintained by vendor ecosystems, not one-off community repos. MCP
ModelOp AI Governance Platform
Use for AI TRiSM compliance and portfolio visibility across agent deployments.
Caveat: enterprise-focused, likely overkill for firms under 100 staff. ModelOp
Tally Forms with conditional logic
Use for client intake automation without custom development. Pair with HubSpot for CRM sync—takes 2 minutes to configure. Tally
Deep Dive | Thesis & Playbook
Most consultancies are stuck in pilot purgatory. Research shows 74% achieve ROI within the first year, but only 10% of use cases scale past proof-of-concept. That 64-point gap isn't a technology problem—it's a capability problem. Firms deploying Microsoft 365 Copilot see productivity gains spread thinly across employees with no visible impact on utilisation or margin. Meanwhile, vertical use cases embedded in specific processes—proposal generation, client reporting, delivery handoffs—remain stuck in the pilot stage despite higher potential for measurable returns.
This week, whilst everyone watched Anthropic donate MCP to the Linux Foundation, the more interesting signal was in the Q4 research from McKinsey, Deloitte, and Google Cloud. Enterprises are willing to wait 12+ months to resolve ROI challenges. They're not rushing. They're building foundations. Here's what separates firms that scale from those that stay stuck.
On Paper
MCP provides a standard way to connect agents to business systems—like USB for AI. 60,000+ open-source projects adopted MCP since the August release. Pre-built connectors now available for Google Drive, Slack, HubSpot, Asana, and Microsoft 365.
Microsoft is launching industry-specific agents through Dynamics 365. Sales Order Agent and Payables Agent show claimed productivity gains of 67% in the Fujitsu case study.
Google Cloud reports 74% of executives achieving ROI within first year. 39% of organisations deploying 10+ agents already. Insurance sector shows 325% adoption increase in regulated environment—8% to 34% in 12 months.
In Practice
Setup requires 5-15 hours of partner or senior consultant time: £500-2,000 opportunity cost before any value delivered. Maintenance needs 2-4 hours monthly when client requirements or processes change—£200-500/month ongoing.
Team adoption friction: 2-4 weeks before consistent usage across delivery teams. Edge cases require manual intervention 10-20% of the time initially. Battery Ventures research shows 42% delta between expected and actual AI usage.
Security vulnerabilities in MCP implementations found: prompt injection, tool permissions, lookalike tools. 49% of practitioners cite data governance as top concern before deployment.
Only 10% of pilots scale past proof-of-concept: Gartner projects 15% of work decisions made autonomously by agents by 2028, up from 0% in 2024—meaning governance frameworks need building now whilst agent footprint is small.
Issues / Backlash
Agent sprawl emerging as firms deploy tools without centralised governance. IT teams struggling with application sprawl—multiple agents doing similar things, abandoned projects, security gaps.
Horizontal tools like Copilot delivering productivity gains that don't show up in utilisation reports or margin analysis. Partner resistance: "yet another tool" adding to stack bloat.
Client data concerns about agent access to sensitive commercial information. Vendor dependency: workflows locked into proprietary platforms, migration painful if pricing changes or tool gets acquired.
My Take (What to do)
Startup: You're capacity-constrained. Partners doing proposal assembly, client reporting, delivery oversight. Start with one workflow that happens 10+ times weekly and eats 30+ minutes each time.
Calculate: (time saved × hourly partner rate × 52 weeks) - (setup cost + annual subscription + 10% contingency). Only proceed if ROI positive within 6 months. Don't build governance frameworks yet—focus on proving value in one high-frequency workflow. Use partner approval for any agent action touching client data.
SMB: Informal processes breaking down, inconsistent delivery across teams, visibility gaps. Your challenge isn't capacity—it's standardisation. Assign one delivery lead to document three high-volume workflows: proposal generation, status reporting, team handoffs. Time them accurately over 2 weeks of real usage. Calculate ROI including maintenance costs. Pilot one automation, measure actual time savings for 4 weeks before expanding.
Now is when you need basic governance: centralised approval for new agents, audit logs, rollback procedures. Appoint one ops team member as "agent owner" managing the portfolio.
Enterprise: Competing with larger firms, margin pressure from client fee expectations, governance requirements slowing delivery. You need enterprise-grade governance without enterprise complexity. Run formal business case with finance team approval. Include change management costs—training, documentation, support. Require 12-month payback minimum. Establish governance framework before deploying: approval workflows, data access controls, audit trails, rollback mechanisms. Create "agent mesh" architecture where agents can be swapped without rebuilding entire workflows—MCP standard helps here.
Focus on vertical use cases that protect margin: proposal automation, client reporting, delivery analytics. Not horizontal tools that spread value thinly.
How to Try (15-minute path)
Pick one workflow you do weekly: Document the steps in a Google Doc. Time yourself doing it once, noting where you wait for information or switch between systems. (5 min)
Calculate baseline cost: weekly time × 52 weeks × your hourly rate (use your target billable rate). This is annual cost of current process. (3 min)
Research one automation option: MCP connector, n8n workflow, or vendor agent. Note setup time in vendor documentation.
Calculate: (annual baseline cost) - (setup hours × hourly rate + annual tool cost + 10% for maintenance).
If positive and payback under 6 months, worth exploring. If negative, this workflow isn't ready. (7 min)
Success metric: Clear yes/no decision on whether this workflow justifies automation investment right now.
Spotlight Tool | MCP
Model Context Protocol (MCP)
Open standard for connecting AI agents to business systems without vendor lock-in. 60,000+ projects adopted since August. Pre-built servers for common platforms. Governed by Linux Foundation with backing from Anthropic, OpenAI, Block, Microsoft, Google, AWS.
→ Google Drive connector
→ Slack integration
→ HubSpot CRM sync
→ Asana project data
→ Microsoft 365 access
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]