IntegrAuth Academy
Identity, security and practical AI — free, from zero. Byte-sized lessons, real diagrams, hands-on labs, zero fluff. Two programs, each with its own exam and certificate.
English is hard? Your browser can translate this page into Tamil, Telugu, Kannada, Malayalam or Hindi. Tech words stay in English so you learn the real names.
How?
- Chrome on phone: ⋮ menu → Translate
- Chrome on computer: right-click → Translate to …
- Edge: the translate icon in the address bar
Picking the right AI tool
Priya's company is trying 30 AI tools. Her manager asks which to keep. Your goal: choose a tool by its job, not its brand.
🧪 Interactive lab — enable JavaScript to play with this one.
- Start with the job, not the brand.
- Check the data terms first.
- Test 2 or 3 tools on 10 of your own real examples.
Start with the job
"Which AI is best?" has no answer. It is like asking which vehicle is best. A bike, a car and a truck each suit a different trip. First, write down 4 things.
- The job. What should be different after?
- The data. Is it public, internal, personal or secret?
- Where it runs. Any cloud, an approved cloud, or only your computer? A cloud is computers run by a company and used over the internet.
- Who runs it. You, your team, or a workflow?
Ten kinds of tools
One product can do several jobs.
| Kind of tool | The job |
|---|---|
| 💬 Chat assistants | Think, draft and explain |
| ⌨️ Coding assistants | Write and fix code |
| 🔎 Research and search | Answer with sources |
| 🎙️ Meetings and notes | Notes and action items from calls |
| 📄 Writing and office | AI inside your documents and email |
| 🎨 Image, video and voice | Make pictures, clips and speech |
| 🔗 Workflow automation | Run steps across your apps |
| 🧩 Agent and app builders | Build chatbots with little code |
| 💻 Local model runners | Run models on your own computer |
| 🔌 Model APIs | Your own code calls a model directly |
Check, then test
Ask about data, cost, and how hard it is to leave. Then run 10 real examples through 2 or 3 tools. Demos are chosen to impress. Your own examples show the truth. This is a small eval.
Now check two tools before you choose one.
🧪 Interactive lab — enable JavaScript to play with this one.
Go deeper: six questions
- 🔒 Data. Where does your text go? How long is it kept? Is it used for training? Free and business plans of one product often differ. See Safe AI at work.
- 💰 Cost. Do you pay per person each month, called per-seat pricing, or per use? Per seat is cheap for 5 people and costly for 500. Pay per use costs little when use is low.
- 🔗 Links to your apps. Does it connect to your documents, chat or code? Does it support MCP, a plug for tools? Can staff sign in with the company login, and can IT remove them in one place?
- 🧳 Leaving. Can you export your data, prompts and workflows? Can you switch the model? How hard it is to leave is called lock-in. Check it before you depend on the tool.
- 🔓 Open or closed. Open source lets you read the code and run it yourself. Closed products are often more polished, with support. But an open-source tool may send your text to a cloud model. Then it is only as private as that model's terms.
- 🩺 Health. Tools can shut down or stop getting updates. With an open license, others can keep the code going. Either way, have a plan to leave.
Go deeper: let the limit decide
| If the limit is… | …look first at |
|---|---|
| No internet, or data must stay on the laptop | A local model runner with an open-weight model |
| Customer personal data | Only a tool your company approved for that data, on business terms. Or one you host yourself. |
| Lots of work, small budget, inside your product | A model API with a small model (Choosing a model). Ask about batch prices. |
| People who don't code must run it | Workflow automation or a no-code builder |
| Every claim needs a source | A research tool, or one that uses only your documents |
| The work lives in documents and email | The office AI of the suite you already use |
Go deeper: examples
Examples as of October 2026. Closed and open tools are mixed, in no order. None is advice.
| 💬 Chat | ChatGPT, Claude, Gemini, Microsoft Copilot, Le Chat, Open WebUI, LibreChat |
| ⌨️ Coding | Claude Code, Cursor, GitHub Copilot, Aider, Cline, Kilo Code, OpenCode, oh-my-pi |
| 🔎 Research | Perplexity, Elicit, deep research in ChatGPT, Gemini and Claude |
| 🎙️ Meetings | Otter.ai, Fireflies.ai, Granola, tools inside Zoom, Teams and Meet, the open-weight Whisper model |
| 📄 Office | Microsoft 365 Copilot, Gemini in Google Workspace, Notion AI, Grammarly |
| 🎨 Media | Midjourney, Adobe Firefly, Runway, ElevenLabs, and the open-weight FLUX and Stable Diffusion. Check the license. Some are for non-commercial use only. |
| 🔗 Workflows | Zapier, Make, Power Automate, n8n, Activepieces, Node-RED |
| 🧩 Builders | Microsoft Copilot Studio, Dify, Langflow |
| 💻 Local | Ollama, llama.cpp, Jan, LM Studio |
| 🔌 APIs | Anthropic, OpenAI, Google, Mistral, Amazon Bedrock, Hugging Face, OpenRouter. vLLM runs models on your own servers. |
Go deeper: Priya's list
Priya listed the 12 jobs her team does again and again. They needed only 5 kinds of tool. For each kind, she ran 10 real examples through 2 tools. She checked the data terms with IT.
The result was a one-page list. It shows which tool fits which job, and what data may go into each. Use of unapproved tools dropped. They were not banned. The approved tools were easy to find, and good at the work.
Staff often use unapproved AI tools on personal accounts. This is shadow AI. It puts work data on consumer terms. Bans often just hide it. Good, approved tools work better. More in AI rules at work.