AI / IT / Power Automate Tool Library

A running library of practical AI, IT and automation tools/techniques worth keeping — sourced mainly from the daily AI/IT digest. Add to this whenever something looks worth revisiting; each entry keeps enough detail to act on later without re-researching it.

Started 27 Sept 2026 for Andy Giles, Herefordshire & Worcestershire ICB.

Tools

Microsoft Copilot rebuild — Home, Code & Autopilot

What it is: Microsoft restructured Copilot into three parts, announced 25 Sept 2026:

  • Home — brings Chat, delegated work, and Office documents together in one place.
  • Code — describe an app, dashboard, automation or workflow in plain language and Copilot builds it.
  • Autopilot — a persistent agent that keeps working on a standing job (e.g. "manage invoices") without you re-prompting it, can sit inside Teams like a colleague, and can be given an identity, direction, memory, computer and workspace of its own.

Link: Microsoft's announcement

Why it matters for you: H&W ICB is a Microsoft 365 shop, so this is the direction Copilot is heading for anyone already licensed. "Code" in particular overlaps with what Power Automate/Power Platform does today — worth watching as a possible alternative or complement for building workflows/automations without leaving the Microsoft ecosystem. Persistent "Autopilot" agents also raise the governance questions covered in the AgentCloak/shadow-AI entries below (auditing, monitoring, who approves what an agent does on a standing basis).

Source: The Neuron, 27 Sept 2026 issue "OpenAI + Anthropic: tens of thousands of AI incidents?".

AgentCloak Desktop

What it is: A free tool from InCountry that automatically swaps sensitive/private values for synthetic placeholders before a prompt is sent to an AI chatbot, then restores the real values in the answer shown on your screen — so the chatbot itself never sees the real data.

Link: incountry.com/products/agentcloak-desktop

Details/caveats:

  • Detection runs on Rampart, a small model published by the US National Design Studio; per its own model card it catches ~98.4% of private terms in Latin-script text but only ~13.7% in non-Latin scripts — treat non-English content with caution.
  • The company's own FAQ says the desktop app currently supports ChatGPT Desktop only, despite marketing referencing wider chatbot support — verify current coverage before relying on it for other tools (e.g. Copilot, Claude).
  • Free; the Chrome extension component is separate from the desktop app.

Why it matters for you: Directly relevant to NHS data-handling constraints around what staff paste into AI chatbots — a possible mitigation if/when people want to use ChatGPT-style tools on commissioning material that contains identifiable information. Worth a proper security/IG review before any real use, not just a personal trial, given it's handling sensitive data by design and the accuracy caveats above.

Source: Neatprompts, 26 Sept 2026 issue "Jev got 39M views. Its founder had 1,500 followers".

Orchesty

What it is: An open-source, self-hostable platform for designing, scheduling and orchestrating asynchronous workflows across connected services and APIs.

Link: orchesty.io

Key features:

  • Use open-source connectors, or generate your own connectors from an API spec with AI
  • Deploy in a managed cloud or fully self-hosted
  • Built-in retries, rate limits, persistence and access controls
  • Pricing: paid (check current tiers on site)

Why it matters for you: A heavier-duty alternative/complement to Power Automate for workflows that need to talk to several external systems reliably (retries, persistence, access control baked in) — worth a look if a commissioning workflow ever outgrows what Power Automate/Power Platform can do cleanly, e.g. chaining multiple system APIs with error handling.

Source: FutureTools newsletter (Matt Wolfe), 25 Sept 2026 issue "YouTube Adds AI".

Gemini Canvas → Google Sheets dashboard

What it is: A built-in Gemini feature (via Google Workspace) that turns a Google Sheet into an interactive, shareable dashboard from a plain-language prompt — no separate BI tool needed.

How to use it:

  1. Open the spreadsheet (e.g. a tracker or project log) in Google Sheets.
  2. Open Gemini in a new tab, click +, select Canvas.
  3. Prompt: "Make me an interactive dashboard using the data from tabs one and two."
  4. Copy the Canvas link to view/share the dashboard outside the chat — colleagues with the link see live updates as you keep editing it via chat.
  5. Pro tip: ask the same Canvas to turn the data into slides too, then export to Google Slides or PDF.

Why it matters for you: A fast, low-effort way to turn an existing tracker/spreadsheet into something presentable and shareable, without building a Power BI report from scratch. Worth checking against H&W ICB's data-governance rules before pointing it at anything containing real patient/commissioning data — treat as a technique for non-sensitive/aggregate data until cleared.

Source: The Rundown AI, 25 Sept 2026 issue "Meta's Connect turns into a Muse takeover".

InSummary's Rosey

What it is: An AI reviewer that checks AI-drafted documents inside Microsoft Word, flagging factual errors and weak spots before they go out.

Link: taaft.co/rosey-r

Why it matters for you: A ready QA layer for anything drafted with AI help — commissioning reports, briefing notes, board papers — that plugs directly into Word rather than requiring a separate app. Worth a trial run on a low-stakes AI-drafted document to see how good its error-catching is before relying on it for anything going out under your name.

Source: There's An AI For That, 27 Sept 2026 issue "Extinct? The Mammoth Will Return".

Free Power Automate HTML Table Styler

What it is: A free tool (by Ellis Karim) that generates clean, styled HTML table markup for use in Power Automate flows — e.g. inside "Send an email (V2)" actions — so you don't have to hand-write CSS/HTML to make an emailed table look presentable.

Link: Free Power Automate HTML Table Styler

Why it matters for you: Power Automate's built-in "Create HTML table" action produces plain, unstyled markup that often looks poor in an email body. This gives ready-made, email-safe HTML/CSS you can drop straight into a flow — a quick win for any Power Automate build that emails a table out (approval summaries, status digests, tracker exports).

Source: Power Platform Weekly, Issue #277 (28 Sept 2026).

Techniques & Prompts

Evals — getting an AI agent to check its own work

What it is: A technique for building a self-check ("eval") step into an AI agent workflow, so the agent verifies its own output against your requirements before handing it back — instead of you catching every mistake manually.

How it works:

  • An eval is just a verifier: does the output meet a requirement you've defined?
  • Put the eval where the agent already looks for instructions:
    • Project-level (e.g. a CLAUDE.md/AGENTS.md file): applies to a kind of work however you ask for it (e.g. "every report gets checked for X before it's shown to me").
    • Skill-level (a SKILL.md): applies to one specific recurring task/procedure.
  • Focus evals on your highest-leverage, most-repeated tasks (80/20 rule) — not everything needs one, since each eval costs extra time/tokens.

Reusable eval prompt (copy/paste):

Evaluate this output against one requirement.
Requirement: [what must be accurate, included, or preserved]
It passes when: [what a good result looks like]
Compare it with: [the original request, source material, or standard it should match]

Give a verdict: pass, needs revision, or not enough information to judge.
For each finding, quote the passage from the output and the source or standard
you're comparing it with, and explain why it meets or misses the requirement.
If it misses, suggest a fix without changing anything yet.

After I approve a fix, check the revised version against the same requirement.

Why it matters for you: If you build any recurring Claude/Copilot-assisted output (commissioning reports, briefing notes, meeting summaries), adding an eval step like this cuts down on manual proofreading/error-catching.

Source: AI Maker (Substack), 24 Sept 2026 — "How to Make Your AI Agent Review Its Work Before You Do".

AI Workflow Automation Finder (prompt template)

What it is: A structured prompt that takes a manual/repetitive process you describe and returns automation opportunities, a prioritised implementation plan, and risk/guardrail flags — built for scoping what's actually worth automating (e.g. with Power Automate) before you build anything.

Full prompt (copy/paste, fill in the brackets):

Act as an Enterprise AI Workflow Automation Finder. Analyze the workflow below and identify where AI, automation, or AI agents could reduce manual work, speed up execution, improve quality, or eliminate repetitive tasks.

Workflow:
[Describe the workflow step by step]

Team/function:
[Sales / Marketing / Finance / HR / Operations / IT / Customer Support / etc.]

People involved:
[Roles and approximate number of people]

Tools currently used:
[Slack, Salesforce, Excel, email, Jira, SAP, Notion, etc.]

Frequency:
[Daily / weekly / monthly / ad hoc]

Time spent:
[Approximate time per task or workflow]

Current pain points:
[Repetitive work, delays, errors, handoffs, bottlenecks, context switching, etc.]

Now analyze the workflow and provide:

1. Automation opportunities
Identify each step that could be:
- Fully automated
- AI-assisted
- Left human-controlled

2. AI workflow designs
For the highest-value opportunities, describe:
Trigger → AI action → Tools/data used → Human approval → Final action

3. Impact
Estimate potential time savings, cost savings, faster turnaround, or quality improvements where the available data supports it.

4. Prioritization
Rank opportunities using:
- Business impact
- Automation feasibility
- Implementation effort
- Risk
- Expected time to value

5. Recommended first automation
Identify the most practical workflow to pilot and explain why.

6. Implementation plan
Give me a simple Pilot → Test → Deploy → Scale roadmap, including required integrations, data, owners, and success metrics.

7. Risks and guardrails
Flag security, privacy, compliance, accuracy, human-approval, and reliability considerations.

8. Executive summary
End with:
- Top 3 automation opportunities
- Expected measurable impact
- Recommended pilot
- Tools/integrations required
- Key risks
- KPIs to track

IMPORTANT:
- Do not assume AI is the right solution for every step.
- Do not invent precise savings, ROI, or productivity figures when inputs are unavailable.
- Clearly label assumptions and estimates.
- Separate measurable benefits from qualitative benefits.
- Challenge automation ideas that add unnecessary complexity.
- Prioritize workflows that are repetitive, high-volume, rules-based, or involve significant manual coordination.
- Preserve human approval for high-risk decisions.
- Highlight missing information that could materially change the recommendation.
- Keep the output concise, practical, and suitable for an enterprise leader deciding what to automate next.

Why it matters for you: Run this against an actual H&W ICB commissioning process (e.g. a recurring report, a waiting-list review, a contract-monitoring cycle) to get a structured, risk-aware first pass at what's worth automating — directly usable as a starting brief for a Power Automate build.

Source: Neatprompts, 25 Sept 2026 issue "The next AI platform could be hanging around your neck".

To add to this library, just say which item(s) from a digest to save — this page gets updated in place rather than duplicated.