Practical team AI field note
Skills, Plugins and Agents: What They Are and Why Your Team Will Want Them
Building a tool once and handing it to your whole team is the moment AI stops being personal. Here is what custom capability actually involves for non-developers.
The core shift
Moving beyond isolated prompting habits creates durable business value.
Individual productivity gains are helpful, but they remain fragile when confined to one employee’s private chat window. Translating your organisation’s standard operating procedures into deployable custom skills, plugins and agents allows every team member to produce consistent, high-quality work.
Across New Zealand and Australia, experienced professionals have spent the past eighteen months developing personal AI routines. They use chat interfaces to summarise board packs, draft emails, draft tender responses and analyse meeting transcripts. These personal habits save real hours every week.
Yet from the organisation’s perspective, those gains remain completely invisible. The company owns none of the process logic, retains no standard documentation, and experiences no shared efficiency. If a key analyst leaves the business on Friday, their prompting routines vanish with them. On Monday morning, the replacement employee starts again from a blank chat box.
This operational fragility marks the boundary between personal AI use and team capability. The step forward is not buying more software licences. It is taking the proven workflows that already work on one desk, packaging them into structured capability, and deploying them so the whole team can use them reliably.
Demystifying skills, plugins and agents
Software vendors frequently surround these tools with dense technical jargon. When you peel back the sales vocabulary, the architecture resolves into three straightforward workplace concepts:
- Custom skills: Pre-packaged instruction sets and reference documents that teach an AI model how to perform a specific recurring task according to your workplace standard.
- Plugins and connectors: Secure technical bridges that permit the model to read from or write to external systems, such as your document storage, CRM database or accounting software.
- Agents: Multi-step workflows where the model uses skills and connectors to complete a broader objective, pausing for human review at defined checkpoints.
As a manager, advisor or business analyst, you do not need to write software code to lead this work. Your job is to define the operational boundaries, document the input requirements and establish what good work looks like. That is process design, not software development.
A custom skill is not software code. It is an agreed standard operating procedure formatted so an AI model can follow it reliably.
Steve Wilson, Zero to AI
How skills function in the real world
To understand how a skill operates inside platforms like Claude or Microsoft 365 Copilot, consider what exists inside the tool package.
A skill is simply a folder containing a text file (often named SKILL.md) and supporting reference files. The text file contains three main parts: a name, a description of the task, and clear instructions. The description field is particularly important. When an employee asks a question, the AI platform reads that description to decide whether this specific skill should take the job.
I saw this design operate clearly in recent work with a regional structural engineering consultancy. The firm had experienced staff spending substantial time reviewing commercial fabrication quotes. Rather than letting forty engineers invent separate prompts, the practice created three dedicated skills:
Estimating skill
Formats project specifications against standard quantity formulas and historic labor bands, producing consistent draft estimates for peer review.
Workshop fabrication skill
Translates project material schedules into clean bill-of-materials tables formatted directly for the workshop planning team.
Finance milestone audit skill
Cross-references engineer progress reports against contractual billing milestones before invoices are prepared for clients.
Data governance and group deployment
A common concern among managers is data security. When you build a custom skill, you must remember that the skill itself does not grant data access rights. It only provides instructions and static template examples.
If your process requires commercial project margins, customer credit histories or private staff records, those figures must remain inside your secure, permissioned databases. The model accesses that information only if the individual user running the skill already has verified permission to view it in systems like Microsoft SharePoint or Entra ID.
Furthermore, platform administration allows leadership to stage access carefully. On Claude, assigning skills or projects to specific groups from the identity provider needs an Enterprise plan, and a Team plan cannot target groups. In Microsoft 365, Copilot works within the permissions the user already has, and groups come from Entra ID. This allows organisations to deploy specialized estimating tools to commercial teams while keeping general administrative tools available to all staff.
When establishing team controls, review the New Zealand responsible AI guidance to make sure your internal data boundaries meet modern workplace expectations.
Practical steps to brief your first skill
To begin converting individual habits into shared team assets this week, follow this four-stage preparation sequence:
- Select one bounded process: Identify a recurring task with consistent inputs, such as client onboarding checklists, weekly variance reports or tender compliance reviews.
- Gather the ground truth: Collect two excellent historic examples of the finished output, along with the source documents used to create them.
- Document the exceptions: Write down the exact conditions where the AI must stop, flag an issue and request manual human review.
- Draft the trigger description: Write a plain two-sentence summary describing exactly when and why an employee should use this capability.
About the author: Steve Wilson is the founder of Zero to AI. With more than 20 years in senior business analysis and experience advising over 400 clients, he helps mid-career professionals across New Zealand and Australia turn technical AI concepts into practical workplace capability.
Build shared capability
Move your team from solo prompting to deployed systems.
Zero to AI shows experienced professionals how to lead AI adoption, govern access sensibly and package practical capability that stays in the organisation.