The Bill Arrives

Tokenmaxxing is over. AI costs are now a CFO problem - and the consultancies that built governance first are about to look very smart.

The Frontier Comes to Everyone

Claude Fable 5 ships, Managed Agents go cron, and the IPO clock starts.

The Coding Shift, the Vatican, and the Pentagon

Anthropic reaches first profit, AI agents delete live code, and the Pope writes 42,000 words about all of it.

The Big Four Have Joined the Room

KPMG deploys Claude to 276,000 staff, PwC follows within the week - and the window for mid-market consultancies to differentiate is now measured in months, not years.

The Deployment Race Is On

OpenAI drops $4 billion on enterprise services, Meta goes proprietary, and the window to build specialised AI delivery capability is narrowing fast.

The Race Nobody Is Talking About

OpenAI ships GPT-5.5 in six weeks, Google locks in $40 billion, and the consultancies pulling ahead aren't winning on model choice - they're winning on workflow design.

The Design Layer Has Entered the Stack

Claude Design ships, Salesforce tears out its UI, and the 20% of firms capturing 74% of AI's value are doing one thing differently.

From Demo to Deployment: Why Most Agent Projects Die Before They Ship

The infrastructure gap is the real bottleneck - and Anthropic just made it someone else's problem.

Agents Are No Longer Experimental. This Week Made That Official.

Managed Agents enters public beta, Muse Spark launches, and Perplexity posts 50% revenue growth after pivoting from search to agents.

The £65K Hire You Don't Need

Claude Cowork enters enterprise via Microsoft, Anthropic leaks its most powerful model yet, and the agentic sales ops playbook is now available to any consultancy with a CRM and £100 a month.

Models Are Becoming Commodities. Workflows Aren't.

OpenClaw proves the model isn't the moat. The firms winning in 2026 are the ones that have operationalised the workflow layer.

The Week AI Stopped Being a Tool

NemoClaw ships, IBM spends $11B on data, Anthropic backs consultancies with $100M — the infrastructure conversation is no longer theoretical.

Your AI Is a Yes-Man. Here's the Proof

Most AI tools are trained to be helpful. The problem is that helpfulness and honesty are not the same thing — and a new independent benchmark this week makes that gap impossible to ignore.

Open Source, Open Wounds: Why the Most-Starred AI Agent Is a Security Liability

OpenClaw hits 95,000 GitHub stars—and 21,000 exposed instances, 341 malicious marketplace skills, and a Belgian national security warning. The real question isn't whether your team is excited about it. It's whether they've already installed it.

OpenCode vs Claude Code: The Cost Question Nobody's Calculating

OpenCode hits 95K stars, Ollama enables local execution—but the real choice isn't which tool is "better", it's whether you've done the maths on what you actually need.

Agents Acquiring Phone Numbers and Security Nightmares

OpenClaw went rogue, Moltbook leaked 1.5M agents' data, and Ollama 0.14 offers a compliance path—if you accept the performance trade-off.

Authorization Bypass: The Governance Failure No One Saw Coming

Agents inherit broad permissions, users access data beyond their roles—traditional security controls don't work for agent-mediated workflows.

Four Interfaces, One Tool, Zero Agreement

Claude Code launches Web, Desktop, IDE, and CLI—consultancies test the wrong one first and abandon before finding what works.

Skills at Enterprise Scale, Governance at Zero

Anthropic ships open standard, consultancies deploy to 350K staff—security teams discover the gap.

The AI predictions that actually matter to consultancies in 2026

Happy New Year From The AiGentic Lab Team

Agent Sprawl vs. Agent Strategy

MCP standardises agent integration, Microsoft ships vertical workflows — but the real bottleneck isn't connectors, it's the discipline to calculate before you automate.

The New AI Stack for Small Business: Simple, Embedded, Everywhere

Copilot launches, Agentforce explodes, BI agents arrive — and SMEs finally get enterprise-grade automation.

Beyond Benchmarks: The Week AI Became Infrastructure

DeepSeek goes open, Gemini Deep Think lands, Trainium3 ships - and sovereign AI becomes a real option.

The Vision Advantage: Gemini 3, GPT-5.1 & the New Agent Reality

What Gemini 3, GPT-5.1 and Scribe v2 mean for your AI roadmap

From Chat to Action: GPT‐5, ChatGPT Apps, Agentforce & Tools