Orphaned Rules - 5 Rules

AI‑assisted tools can turn rough ideas into working demos in hours instead of weeks. They help you scaffold codebases, generate UI from prompts or designs, and wire up data so you can validate scope and risk with clients quickly.

Tooling Options

Here is a list of AI code generation tools:

• Github Spark (Copilot Pro+ only)

  GitHub Spark is an AI‑powered app builder that turns natural language instructions into full‑stack TypeScript/React apps, complete with live preview and GitHub repo integration. It’s tightly integrated with GitHub Copilot and Codespaces, making it easy to go from idea → prototype → hosted demo quickly. Spark is perfect for building end‑to‑end demos directly within the GitHub ecosystem.

• Base44

  Base44 focuses on full‑stack scaffolding. By simply describing your app, it spins up CRUD operations, authentication, forms, and basic data flows. It’s particularly helpful when you need a working skeleton to show user journeys or data interactions during client presentations.

• v0

  v0 by Vercel is a UI‑focused generator that outputs production‑ready React and Tailwind components. It’s a great option when you need to iterate on design directions quickly or want to build out front‑end layouts that work seamlessly with Next.js projects.

• Firebase Studio

  Firebase Studio leverages AI to help you scaffold backends, define Firestore data models, generate security rules, and create sample data. It’s ideal when your prototype needs authentication, cloud functions, and real‑time data syncing without heavy backend setup.

• Lovable

  Lovable focused on responsive design. It helps you generate front-end and full-stack applications that adapt seamlessly from desktop to mobile. You can start from a prompt, an image, or directly from a Figma file. It's especially useful when you need polished, responsive layouts that work across screen sizes out of the box.

• Bolt.new

  Bolt.new supports multiple frameworks beyond React, such as Vue, Svelte, and Angular. It offers terminal access for running specific commands and supports integrated deployment, so you can go from prompt (or even an image) to a live site in minutes. Ideal for quick prototyping in non-React stacks or showcasing cross-framework concepts.

• Anima

  Anima specializes in turning high-fidelity designs into near pixel-perfect React, HTML, and CSS code. It integrates directly with tools like Figma, Sketch, and Adobe XD via plugins, making it easy to export real, production-grade code from your design files. It's a great choice when visual accuracy and front-end alignment with design specs are a top priority in your prototypes.

• Uizard

  Uizard acts like a pseudo-designer, allowing you to quickly generate multi-page UI designs from prompts or even screenshots. It supports exporting to code, making it ideal for rapid prototyping or client-facing mockups without needing full design expertise. It's especially handy for quickly visualizing product ideas or user flows in minutes.

✅ Best Use Cases for AI Tools

Rapid prototyping and design exploration

Non-technical team members can use screenshots, hand-drawn wireframes, or Figma files to create functional prototypes. These tools allow quick iteration, fast feedback, and better alignment across teams early in the design process.

Kick starting new projects

Use AI-generated code as a base to accelerate development. Many tools produce clean, component-based layouts that follow design principles and give developers a working foundation — helping teams skip repetitive boilerplate and focus on core features.

Going from idea to deployment

Some tools like v0 and base44 can take a project from wireframe to a deployed demo with minimal effort. This helps teams validate concepts with stakeholders, collect feedback, and iterate fast — bridging the gap between idea and implementation.

Replicating and reusing UI patterns

AI tools like v0 or Anima are great for extracting patterns from reference sites — e.g., navigation, pricing tables, or forms — and turning them into working components. These can be integrated into your design system, refined, and styled to meet brand or accessibility standards.

❌ What to Avoid When Using AI Tools

AI tools are great for prototypes, but they do not replace good software engineering. Here are common mistakes to avoid:

Treating prototypes as production code

AI-generated code is built for speed, not safety or scalability. It often lacks error handling, validation, and test coverage. Shipping this code directly to production can lead to security issues, crashes, and long-term maintenance problems. Always treat prototypes as drafts — they must be reviewed and hardened before deployment.

Skipping human review

AI can generate structured code, but it doesn’t understand your business logic or security standards. That’s why every AI-generated change should be reviewed — especially pull requests or multi-file edits. Never auto-merge AI output. A human eye helps catch logic bugs, performance issues, and unsafe assumptions.

Uploading real client data

Do not paste real or sensitive client data into prompts or online AI tools. Most tools process data in the cloud, and unless there’s a verified agreement in place, you risk a data breach or compliance violation. Always use fake or anonymized data during prototyping.

Ignoring licensing and attribution

Some generated content may be derived from licensed or attributed sources. Before using AI-generated code or media in a project, always verify its origin and license. This is especially important if your prototype is going to production or reused in commercial contexts.

Example of prompt and the result

AI‑assisted tools can turn rough ideas into working demos in hours instead of weeks. They help you scaffold codebases, generate UI from prompts or designs, and wire up data so you can validate scope and risk with clients quickly.

Tooling Options

Here is a list of AI code generation tools:

• Github Spark (Copilot Pro+ only)

  GitHub Spark is an AI‑powered app builder that turns natural language instructions into full‑stack TypeScript/React apps, complete with live preview and GitHub repo integration. It’s tightly integrated with GitHub Copilot and Codespaces, making it easy to go from idea → prototype → hosted demo quickly. Spark is perfect for building end‑to‑end demos directly within the GitHub ecosystem.

• Base44

  Base44 focuses on full‑stack scaffolding. By simply describing your app, it spins up CRUD operations, authentication, forms, and basic data flows. It’s particularly helpful when you need a working skeleton to show user journeys or data interactions during client presentations.

• v0

  v0 by Vercel is a UI‑focused generator that outputs production‑ready React and Tailwind components. It’s a great option when you need to iterate on design directions quickly or want to build out front‑end layouts that work seamlessly with Next.js projects.

• Firebase Studio

  Firebase Studio leverages AI to help you scaffold backends, define Firestore data models, generate security rules, and create sample data. It’s ideal when your prototype needs authentication, cloud functions, and real‑time data syncing without heavy backend setup.

• Lovable

  Lovable focused on responsive design. It helps you generate front-end and full-stack applications that adapt seamlessly from desktop to mobile. You can start from a prompt, an image, or directly from a Figma file. It's especially useful when you need polished, responsive layouts that work across screen sizes out of the box.

• Bolt.new

  Bolt.new supports multiple frameworks beyond React, such as Vue, Svelte, and Angular. It offers terminal access for running specific commands and supports integrated deployment, so you can go from prompt (or even an image) to a live site in minutes. Ideal for quick prototyping in non-React stacks or showcasing cross-framework concepts.

• Anima

  Anima specializes in turning high-fidelity designs into near pixel-perfect React, HTML, and CSS code. It integrates directly with tools like Figma, Sketch, and Adobe XD via plugins, making it easy to export real, production-grade code from your design files. It's a great choice when visual accuracy and front-end alignment with design specs are a top priority in your prototypes.

• Uizard

  Uizard acts like a pseudo-designer, allowing you to quickly generate multi-page UI designs from prompts or even screenshots. It supports exporting to code, making it ideal for rapid prototyping or client-facing mockups without needing full design expertise. It's especially handy for quickly visualizing product ideas or user flows in minutes.

✅ Best Use Cases for AI Tools

Rapid prototyping and design exploration

Non-technical team members can use screenshots, hand-drawn wireframes, or Figma files to create functional prototypes. These tools allow quick iteration, fast feedback, and better alignment across teams early in the design process.

Kick starting new projects

Use AI-generated code as a base to accelerate development. Many tools produce clean, component-based layouts that follow design principles and give developers a working foundation — helping teams skip repetitive boilerplate and focus on core features.

Going from idea to deployment

Some tools like v0 and base44 can take a project from wireframe to a deployed demo with minimal effort. This helps teams validate concepts with stakeholders, collect feedback, and iterate fast — bridging the gap between idea and implementation.

Replicating and reusing UI patterns

AI tools like v0 or Anima are great for extracting patterns from reference sites — e.g., navigation, pricing tables, or forms — and turning them into working components. These can be integrated into your design system, refined, and styled to meet brand or accessibility standards.

❌ What to Avoid When Using AI Tools

AI tools are great for prototypes, but they do not replace good software engineering. Here are common mistakes to avoid:

Treating prototypes as production code

AI-generated code is built for speed, not safety or scalability. It often lacks error handling, validation, and test coverage. Shipping this code directly to production can lead to security issues, crashes, and long-term maintenance problems. Always treat prototypes as drafts — they must be reviewed and hardened before deployment.

Skipping human review

AI can generate structured code, but it doesn’t understand your business logic or security standards. That’s why every AI-generated change should be reviewed — especially pull requests or multi-file edits. Never auto-merge AI output. A human eye helps catch logic bugs, performance issues, and unsafe assumptions.

Uploading real client data

Do not paste real or sensitive client data into prompts or online AI tools. Most tools process data in the cloud, and unless there’s a verified agreement in place, you risk a data breach or compliance violation. Always use fake or anonymized data during prototyping.

Ignoring licensing and attribution

Some generated content may be derived from licensed or attributed sources. Before using AI-generated code or media in a project, always verify its origin and license. This is especially important if your prototype is going to production or reused in commercial contexts.

Example of prompt and the result

“Your loan is approved under Section 42 of the Banking Act 2025.”
One problem: there is no Section 42.

That single hallucination triggered a regulator investigation and a six-figure penalty. In high-stakes domains like finance, healthcare, legal and compliance zero-error tolerance is the rule. Your assistant must always ground its answers in real, verifiable evidence.

1 – Why high-stakes domains punish guesswork

• Regulatory fines, licence suspensions, lawsuits
• Patient harm or misdiagnosis
• Massive reputational damage and loss of trust

When the error budget is effectively 0%, traditional “chat style” LLMs are not enough.

2 – The three-layer defense against hallucination

2.1 Retrieval-Augmented Generation (RAG)

• What it does – Pulls fresh text from authoritative sources (regulations, peer-reviewed papers, SOPs) before answering.
• Win – Grounds every claim in evidence; supports “latest version” answers.
• Risk – Garbage in, garbage out. A bad retriever seeds bad context.

2.2 Guardrail filter

• What it does – Post-processes the draft answer. Blocks responses that:

  • lack citations
  • creep into forbidden advice (medical, legal)
  • include blanket “always/never” claims
• Win – Catches risky output before it reaches the user.
• Risk – Over-filtering if rules are too broad or vague.

2.3 Question sanitizer

• What it does – Rewrites the user prompt, removing ambiguity and hidden assumptions so retrieval hits the right documents.
• Win – Sharper queries ⇒ cleaner answers.
• Risk – Requires strong NLU to keep the chat natural.

Rule of thumb: Use all three layers. One patch isn’t enough.

3 – Reference architecture

1. Vector store & embeddings – Pick models that benchmark well on MTEB; keep the DB pluggable (FAISS, Pinecone, Azure Cognitive Search).
2. Retriever tuning – Measure recall@k, MRR, NDCG; test different chunk sizes and hybrid search.
3. Foundation model & versioning – Record the model hash in every call; monitor LiveBench for regressions.
4. Guardrails – Combine rule-based (regex) and model-based tools (OpenAI Guardrails, Nvidia Nemotron Guardrails).
5. Audit logging – Append-only logs of user prompt, retrieval IDs, model version, guardrail outcome.

4 – Measurement is mandatory 🧪

Track from Day 0:

• Exact-answer accuracy (human-graded)
• Citation coverage (every claim cited)
• Compliance errors (dosage mismatch, policy breach)
• Hallucination rate (uncited claims)
• Retrieval miss rate (index drift, ACL failures)

5 – Scaling safely

Auto-fallback to a human expert if any metric dips below threshold.

6 – Domain experts are non-negotiable

• Source curation – SMEs tag “gold” paragraphs; retriever ignores the rest.
• Prompt reviews – Experts catch edge cases outsiders miss.
• Error triage – Every failure labeled with why it failed (retrieval miss, guardrail gap, model hallucination).

Treat specialists as co-developers, not QA afterthoughts.

7 – Key takeaways

• Layer it on – RAG + sanitization + guardrails deliver the most robust defense.
• Measure everything – Strict, automated metrics keep you honest.
• Log & secure by default – ACLs, encryption, append-only audit trails.
• Scale with care – Stay human-in-the-loop until the data proves otherwise.

Nail these practices and you’ll move from a flashy demo to a production-grade AI assistant that never makes up the rules or facts.

If you've ever missed important emails from an Outlook Group, despite being a member, it's likely you weren't following the group in your inbox. This can be especially confusing when those messages don't show up in search results or your inbox, even though you technically "have access."

When you're added to an Outlook Group, you don't automatically receive group messages in your inbox - they only show up in the group's shared mailbox. If you want those messages to behave like regular emails (appear in your inbox and show up in search), you need to explicitly follow the group.

How to follow an Outlook Group in your inbox

1. Open Outlook and go to the left-hand navigation pane.
2. Under Groups, find the group you're a member of.
3. Click on the group (e.g. "SSW TinaCloud").
4. At the top-right of the message panel, click the dropdown next to Following in inbox.
5. Select Receive all email and events.

Admins: Set this as the default for new members

If you're an admin of the group, you can set this to be the default so all members automatically follow the group in their inbox when they're added. This prevents confusion and ensures new members don't miss important updates.

1. Go to the group's Members tab.
2. Click the pencil icon to edit the group settings.
3. Check the Subscription box:
   "Members will receive all group conversations and events in their inboxes."

Why this matters

• Search works properly - If you don't follow the group, messages won't appear in your normal email search.
• Consistency - You'll get emails the same way you get all other mail, making it easier to keep up with group activity.
• Inbox Zero still works - You can still manage and triage these messages with your usual workflow.

If you're missing group emails, this setting should be your first check.

Development for SharePoint is very different depending upon whether you are online or using On-Premises SharePoint.

For SharePoint Online

All you need is VSCode – all modern customizations are doing using the SharePoint Framework (SPFx).

💡 Best practice is to isolate your code as much as possible during development/staging. Unfortunately it's not easy to come up with a full-blown dedicated Test Tenant.
As a minimum, make sure you test all your custom code in a dedicated Site Collection to avoid as much side effects as possible.

See our rule to modern SharePoint Development.

For SharePoint 2016, 2019 or SE

1. It's very important to correctly setup a SharePoint environment for development. Correctly configured, this will save you a lot of trouble later on
2. From time to time, you can seriously damage a SharePoint installation during development and it is best not to install SharePoint on your everyday working machine. Additionally, when you start a new SharePoint project you don't want to carry all the luggage from a previous customization that could potentially affect your new project
3. Virtual machines can be fired up and shut down easily
4. Virtual machines can be relocated on a different server and thus it doesn't waste developers' own computer resources
5. Virtual machines can be copied and brought to a client for demonstration
6. Very easy for someone to quickly create a new SharePoint server to quickly test or experiment with SharePoint
7. Bad - There might be more work required to activate additional servers. SharePoint Farms are a lot of work. E.g. Search Server VMs