The models are no longer the bottleneck
The model is rarely the reason a project fails now. Every major lab has shipped something capable at multi-step reasoning, tool use, and long-context work.
What's changed is what sits around it — orchestration, memory, evaluation, the plumbing into real business systems. That's where reliable outcomes actually come from.
Agents are looking like the hype promised
"Agentic AI" was mostly aspirational through 2024 and 2025. In 2026 it's real.
OpenAI launched Frontier in February. They're calling them "AI coworkers" — agents you onboard like an employee, with shared context, scoped permissions, training and feedback. Frontier plugs into your CRM, data warehouse, ticketing tools and internal apps so agents understand how work actually gets done.
Anthropic's Claude Cowork went enterprise-GA in April. It sits on your desktop, reads and writes to local files, and handles multi-step tasks end-to-end. Give it a goal, come back to a deliverable.
Both are shipping role-specific capability. Anthropic has plug-ins for finance, legal, HR and marketing. OpenAI brought in BCG, McKinsey, Accenture and Capgemini to help enterprises redesign workflows around this.
The shift: agents aren't stuck in a chat window anymore. They work in the same folders and apps a human teammate would, with scheduled tasks, persistent memory, and the ability to hand work back when judgment is needed. Meaningfully different from the chatbots most SMEs are using today.
Security has grown up
Governance is the bit that's changed most quietly — and it's what unlocks adoption for serious businesses.
Both platforms ship with it built in. Each agent has its own identity. Permissions scope to the task — read-only to one SharePoint folder, write access to one Notion page, no over-provisioning. Every action is auditable. Cowork emits OpenTelemetry events for every tool call and file change, straight into Splunk or whatever SIEM you already run.
Connectors can be read-only or write-only, set once from the admin console. Agent activity sits in the same dashboards IT already uses.
If you've tried to get an AI pilot past a cautious IT team, this is the thing. The answer used to be "come back when it's more mature." It's more mature.
The gap for Australian SMEs has changed shape
Adoption isn't the story anymore.
In the US, around 58% of small businesses use generative AI regularly, up from 40% a year ago. Australia sits between 64% and 84% depending on the survey.
But headline usage is the wrong number. The more useful finding from the 2026 State of AI Adoption in Australian SMBs: only about 5% of SMEs are "fully enabled" — AI embedded into core workflows, not used ad-hoc for emails and meeting summaries. Intermediate adopters see a 45% profitability uplift. Fully enabled businesses see 111%. (The skills gap data from 2025 pointed to exactly this pattern emerging.)
The gap isn't "do you know what ChatGPT is" anymore. It's "have you redesigned a workflow end-to-end around what these tools can do." Very different conversation than twelve months ago.
What this means going forward
For the last two years, the right first step for most SMEs was education, exposure, a few safe pilots. That's still where some need to start. But the frontier of useful advisory work is shifting.
Strategy alone isn't going to cut it for much longer. Most SMEs don't need more frameworks. They need someone to sit with a specific workflow — quoting, client onboarding, weekly reporting, procurement — and help design an agentic version of it. Scope the permissions. Pick the platform. Build the evaluation loop. Train the team. Measure what comes back.
This is where we're focusing at The AI Guides. Practical agentic workflow advisory for Australian SMEs — the work that moves a business from the 80% who use AI to the 5% who have it embedded.
If that's where you're heading, have a chat.
Cheers,
Patrick
---
Relevant Insights
- News in AI – January 2026 — The prior edition: AI moving from experimentation to deployment and what Australia's National AI Plan means for SMEs
- Executive Guide to AI Governance for SMEs — A simple governance setup that unlocks responsible AI use across your business
- AI Implementation Guide — A step-by-step framework for moving from pilot to embedded AI workflows
- AI Strategy Essentials for SMEs — How to build a practical AI strategy for your business
---
About the Author
Patrick is co-founder of The AI Guides, bringing a decade of strategy consulting experience to help Australian SMEs adopt AI with confidence. Based in Sydney, he specialises in practical AI strategy, executive training, and building team capability.
About The AI Guides
The AI Guides helps Australian SMEs navigate AI adoption with confidence. We provide expert AI strategy, executive and team training, and implementation support tailored to your business needs. Founded by two Sydney-based strategy and digital transformation professionals, we serve as your trusted guides through the evolving AI landscape.