Knowmium toolkit for change

Make something just for you.
Start smaller than the hype.

A practical guide to making a personal app for fun, learning the basics of Microsoft Copilot agents, and following the few frontier-AI sources worth your attention. No pitch deck. No business case. Just curiosity.

Last reviewed: 15 September 2026 For curious people, after work Make it small. Make it yours.

Start with the human work

Four resources to make the work sharper before the tool gets bigger.

AI changes the leverage. It does not remove the need for good judgment, a clear audience, a testable claim, or an honest view of what could go wrong. Start here before choosing a platform or building an agent.

Project thinking tool

AI Project Pre-Mortem

Pressure-test an idea before you put serious time behind it. The prompt separates what you know, what you assume, and what you need to discover—then turns failure into useful action.

Run the pre-mortem →

Persuasion toolkit

Persuasive Narrative Prompts

Build an opening narrative that helps an audience see why a challenge matters and what action is now worth taking. Works in Copilot, ChatGPT, Claude, Gemini, and Perplexity.

Explore the prompts →

Presentation library

Slides That Decide

Turn data, evidence, and ideas into consulting-quality presentations. Use the Power Prompts to build the narrative, stress-test the logic, and prepare for the room.

Open the prompt library →

For the people who build people

Uplift L&D

Guides, tools, case studies, an AI readiness check, and research for trainers, facilitators, HR leaders, and L&D professionals using AI thoughtfully in real practice.

Explore Uplift L&D →

01 · Make an app

Make a small app for your own life, just to see what happens.

The point is not to build a company or automate your job. It is to make something playful, useful, or oddly specific for yourself: a trip planner, a book-club picker, a dinner-decider, a hobby tracker, a household game, or a tiny site for an idea you cannot stop thinking about.

Start with a tiny truth: choose something you already care about, then make the smallest version that could be fun to show one friend. The first version does not need a business model, a workflow, or permission from anyone.

Get started: make a quick personal app

Pick the tool that makes you want to begin tonight. Every option below can help you turn a sentence into a working web app; use only information you are happy to put into a personal experiment.

Easy personal toolFree plan available · Starter from $16/mo annual

Base44

Try it for: a personal dashboard, shared household helper, habit experiment, travel companion, or a small tool that makes a hobby easier. Describe the thing you wish existed, then keep refining it in plain language.

Make something with Base44 →
Polished & playfulFree plan available · Pro from $25/month

Lovable

Try it for: a beautiful personal web app, a birthday-game site, a creative side-project, a reading companion, or an idea that deserves a little visual personality. Talk to it as you would a smart creative collaborator.

Make something with Lovable →
Experimenter’s routeFree tier available

Google AI Studio Build

Try it for: a more imaginative web or Android experiment—especially one where an AI feature is part of the fun. It is a brilliant place to play with what is possible, then inspect and export what it creates.

Make something with AI Studio →
Keep the experiment personal: do not paste work, client, employee, confidential, financial, or sensitive personal information into a free AI builder. Start with an idea, a few made-up examples, and data you would be comfortable sharing with a friend.

02 · Learn Copilot agents

Learn the small, useful version before the ambitious version.

The first goal is not an autonomous agent. It is a trustworthy information assistant that knows its source boundary, its limits, and when to hand a question back to a human.

Scope for this guide: Microsoft 365 Copilot, Agent Builder, and Microsoft Copilot Studio. GitHub Copilot, code-first agents, external integrations, MCP, autonomous triggers, and multi-agent systems are deliberately outside the first learning path.

Build an agent step by step

  1. Agent in a Day: complete Modules 1–2 only. They cover Agent Builder and a grounded Copilot Studio agent.
  2. Copilot Studio quickstart: use the create–test–improve loop under supervision. Do not use its demo-site sharing pattern for HR material.
  3. Write agent instructions: connect the instruction to the knowledge and capabilities actually configured.
  4. Agent Academy Recruit: use Standard-harness Missions 00, 01, 02, 04, and 06 selectively.
  5. About agent evaluation: create a human-reviewed test set before a named-user pilot.

Step 01

Learn & practise

Use only synthetic or non-sensitive examples. Make human review a habit before the build starts.

Step 02

Choose one question

Pick a small information problem, such as policy navigation or new-hire support—not a people decision.

Step 03

Curate the sources

Use a small, owned, current SharePoint source set with a named policy owner and a review date.

Step 04

Test, then release narrowly

Test access, source grounding, edge cases, and escalation. Share only with a named, authenticated pilot group.

A good first agent: a read-only HR Policy Navigator or New-Hire Support assistant. It should point people to approved sources, state when it cannot answer, and route exceptions, personal determinations, benefits questions, or legal/medical/tax matters to a human owner. It should not make employment or eligibility decisions.

03 · Follow AI signal

Follow fewer sources. Give every source a job.

Do not build an inbox that confuses product announcements with independent evidence, a preprint with peer review, or a market opinion with a filing. This stack is designed for frontier awareness with a finance and operating lens.

Technical frontier

Interconnects is the standing technical interpreter for model reviews, open models, and post-training. Use Simon Willison when agent security, prompt injection, tool use, or practical model behaviour becomes relevant.

Explore Interconnects →
A 30-minute weekly habit: read Reuters, The Batch, and One Useful Thing; capture no more than three developments; then open a specialist or primary source only when a development could change workforce practice, operating workflows, technology spend, AI infrastructure, security exposure, or a diligence question.

Your next small thing

Make one small thing.
See what it teaches you.

The fastest way to understand this technology is to play with it outside the meeting, outside the brief, and outside the pressure to make it impressive. Choose an idea with a personal stake. Make the first version in an evening. Learn from what happens next.

Tonight

Choose a small delight

Pick a recurring annoyance, a hobby, a joke with friends, or a curiosity you want to explore. Keep it tiny enough to finish badly and still enjoy.

Tonight

Describe the feeling

Tell the builder who it is for, what it should help with, and the mood you want. Then ask it to make one working screen before anything else.

This week

Let someone play

Send the link to one friend. Ask where they smiled, where they got stuck, and what they wished it did next. Change one thing.

If it grows

Run the pre-mortem

When the experiment begins to matter, ask what could disappoint, confuse, or break. Keep the human decision with you.

Build the capability. Keep the human in charge.

Tools, prices, free tiers, licenses, interfaces, and publishing rules change. Confirm current terms before you pay or share. This guide is for personal learning and experimentation, not a security approval, legal opinion, or a reason to put work information into a personal AI tool.

Research sources are linked on each resource. Core product and learning claims were checked against official vendor and Microsoft documentation on 15 September 2026. This embed contains no input fields, scripts, tracking, data capture, storage, or external calls.