Level 01
Personal tools
Ask, draft and summarise. Useful, but isolated and easy to lose.
Works.
AI implementation company
AI implementation company
Currents AI Roundtable
Lessons from the field
Lessons from the field
90% of companies
are still at Level 1.
In the companies we see, AI is still mostly isolated chat and scattered experiments. The goal is to turn useful work into reliable ways of working that the whole team can use.
Level 02
Personal workflows
One person repeats a reliable process around their own work.
Level 03
Team workflows
The process is shared, uses real inputs and has a clear owner.
Level 04
Cross-functional
Connected workflows move across teams and improve with use.
Raise the floor. Turn what your best AI users learn into shared ways of working. Leadership has to drive adoption visibly from the top.
Works. // Lesson 0101 / 05
Lesson 02 · Start with the work
Map the work before you add AI.
Start with a process the team actually runs. Map it from trigger to outcome, then decide what to automate, assist or keep human.
01 · Trigger
What starts the work?→
02 · Steps
Who does what, where?→
03 · Outcome
What does done mean?Automate
No judgement
Transcription, re-keying, lookup, formatting and routing.
Assist
Human decides
AI drafts, summarises or surfaces options. A person applies judgement.
Keep
Human-owned
Relationships, physical work, legal sign-off and final accountability.
First pilot
Frequent + painful + measurable
Build the smallest useful version. Run it live. Then scale or stop.
Works. // Lesson 0202 / 05
Lesson 03 · Build the operating layer
A working AI system needs four layers.
The tool is not the system. Build the stack underneath it and keep the models replaceable.
01 · Architecture
Where it lives
Models, tools, data and permissions wired into the business.
Model-agnostic by design
02 · Context
What it needs
The smallest set of current business facts required for this job.
Relevant beats exhaustive
03 · Brain
What compounds
Context organised, maintained and shared as company memory.
Decisions get written back
04 · Skills
How work runs
Executable SOPs that make a proven process repeatable.
Skills come last
Models will change under you. The brain and skills should compound.
Works. // Lesson 0303 / 05
Lesson 04 · Build the capability
Enablement is the blocker to AI value.
The model is one layer. The hard part is helping people use it in real work and spreading what works.
01 · Find
Choose the right work
Map real workflows, find the valuable problems and give each opportunity an owner.
02 · Prove
Prove the value
Build on live inputs, measure the result and fix what breaks before scaling.
03 · Spread
Scale what works
Train the team, set ownership and guardrails, and turn proven systems into shared capability.
Anthropic and OpenAI have put $5.5B behind dedicated companies that help enterprises turn AI into working systems.
Anthropic · $1.5B
Ode with AnthropicOpenAI · $4B
OpenAI Deployment CompanyThe model can be bought. Enablement has to be built.
Works. // Lesson 0404 / 05
Works services
A clear path from AI ideas to company capability.
Start with evidence. Install the operating layer. Build only what the business needs. Then keep it current.
01 · Audit
Find the work
1–4 weeks
Map the workflows, rank the opportunities and leave with a practical roadmap.
02 · Embed
Install the system
90 days
Build the company brain, ship 6–10 workflows and train an internal owner.
03 · Build
Add what is missing
When needed
Add custom infrastructure only where the existing stack cannot do the job.
04 · Optimize
Keep it current
Ongoing
Maintain the brain, add new skills and improve the workflows already in use.
Works stays until the capability works. Then your team owns it.workshq.com.au
Works. // How we operate05 / 05