Good work to support.
- Finding and summarising relevant information.
- Classifying documents and drafting structured outputs.
- Preparing research with traceable sources.
- Helping people navigate a defined workflow.
What I do / AI applications & automationFrom manual work
From manual work
to intelligent,
connected workflows.
Find where AI supports a real task, then design the information, tools and controls around it.
Explore an opportunityUseful context.
Clear boundaries. Human judgement.
Not every problem
needs an agent.
Choose the simplest approach that fits the work. Select an approach to see where it belongs.
Automation
Known rules. Predictable steps.
When the conditions and the action are clear, a straightforward rule or integration may do the job well.
A defined triggerA known ruleA predictable action
Illustrative exampleA submitted enquiry creates a record and notifies its owner.
Start here when the process does not need interpretation.AI-assisted workflow
A defined process. Help with interpretation.
Use a model for a bounded language or document task inside a process whose next steps are already defined.
Relevant inputA bounded model taskReview & continue
Illustrative exampleSummarise an incoming document, attach the source and ask a person to check the draft.
Evaluate output quality, missing information and failure handling.Agentic system
Bounded decisions about what to do next.
A model uses permitted tools, looks at the results and chooses further steps within explicit limits.
A goal & constraintsTool use ↔ resultsStop or ask a person
Illustrative exampleGather information for a meeting, notice a missing record and ask for clarification before continuing.
More autonomy means more work on reliability, control and security.A model is one part
of a useful system.
Select a layer. This is a conceptual view of the design questions, not a blueprint every AI product must follow.
Working with language
A model can draft, classify, summarise or reason over an input. It needs a clear job and a way to assess whether the output is useful.
The information that matters
Relevant documents, current records and the purpose of the task. Choose what the model can see, keep sources traceable and surface gaps.
Access to the right systems
Defined connections let a system retrieve information or propose an action. Access should be limited to the task and the user’s permissions.
Boundaries around the work
The process decides what happens next, when to retry, when to stop and when to ask a person. More autonomy calls for stronger checks.
Judgement stays visible
Important decisions, uncertain outputs and consequential actions need an appropriate review point and a clear owner.
From a request
to a useful briefing.
An example of how connected tools and relevant information can change the experience.
“Prepare me for tomorrow’s customer meeting.”
Illustrative exampleA standalone chatbot
“Here’s a suggested agenda and a few preparation tips.”
Useful advice. You still find the information.A connected workflow
The brief is prepared from the context you’ve permitted it to use.
Select a step to see the work behind it.Tomorrow, 10:00 · Acme customer review
Find tomorrow’s customer meeting in the calendar you have allowed the system to read.
Attendees identified from the invitation. If there are two plausible meetings, ask which one.
Permitted sources onlyReview before action
Account context, brought together
Retrieve relevant records, recent exchanges and documents within the user’s permissions.
Current objectives, the last conversation and open actions. Missing or conflicting records stay visible.
Permitted sources onlyReview before action
A source-backed view of what changed
Look for relevant company developments. Keep sources attached and distinguish evidence from interpretation.
Recent updates sit beside their source links. Uncertain matches are flagged rather than treated as facts.
Permitted sources onlyReview before action
Your meeting brief
Assemble a concise briefing around the purpose of the meeting, with open questions and gaps.
Account summary · previous commitments · suggested discussion points · questions to clarify.
Permitted sources onlyReview before action
Ready for your judgement
Check the evidence, correct the gaps and decide what to use. Nothing is emailed or committed on your behalf.
The system gathers and prepares. You decide what matters, what is correct and what happens next.
Permitted sources onlyReview before action
A design illustration, not a client delivery or a live integration. No tools are connected and no data is sent.
Keep judgement
in the right hands.
- Define permitted data and tool access.
- Check uncertain or consequential outputs.
- Plan for failure, escalation and stopping.
- Keep an owner for important decisions and actions.
Sometimes you need a clearer process or a simpler piece of software. That is a useful answer too.
A practical place
to start.
Map a real workflow, find a bounded opportunity and prototype it with appropriate data. Test usefulness, failure cases and the review experience before deciding how far to take it.
I connect product design with technical decisions and AI-assisted implementation. Responsibilities, data handling and release criteria need to be explicit.
See a real publishing workflowHave a problem
worth solving?
Whether you’re exploring an idea, improving a process or figuring out where AI fits, I’d be happy to talk.