Understand AI.
Stay in control.
AI is no longer just chat. Today's systems can reason through harder tasks, work across text, images and audio, connect to tools, and take multi-step actions. The AI Compass explains what matters — without hype or unnecessary jargon.
Reasoning modelsAI agentsSafety & provenanceAI changed quickly. Your mental model should too.
Four useful starting points for the way people actually use AI now.
What are AI agents?
How models use tools, browse, remember state and complete multi-step work — plus the risks.
02 / GUIDEPrompting in 2026
Move beyond prompt tricks. Learn context engineering, source grounding and reusable task briefs.
03 / PROTECTStay safe with AI
Prompt injection, private data, deepfake scams, dangerous permissions and verification habits.
04 / RULESWhat rules apply?
A practical map of the EU AI Act, NIST's risk framework, provenance standards and copyright.
Four shifts worth understanding
These changes explain why AI feels different from the chatbots of only a few years ago.
Reasoning models
Some models spend extra computation on difficult problems. They can be stronger at planning, coding and analysis — but still need checking.
AI agents
Models can now call tools and execute workflows. Capability rises, but so does the importance of permissions, logs and approvals.
Connected AI
Standards such as MCP make it easier to connect models to apps and data. Every connection also widens the security boundary.
Synthetic media rules
AI-generated media is meeting stronger transparency and provenance requirements, including EU obligations that began applying in August 2026.
Learn the part of AI you actually need
You do not need to become a machine-learning engineer to use AI intelligently.
A clearer map of modern AI
Short enough to read, deep enough to be useful.
AI in Plain English
The minimum useful mental model: machine learning, generative AI, multimodal systems, reasoning and the limits that still matter.
LLMs & Modern Models
Tokens, context windows, RAG, tool use, reasoning models, fine-tuning and why a model can be capable without being reliably factual.
AI Agents
Understand the agent loop, tools, MCP, permissions, failure modes and a safer read → draft → approve → act pattern.
Prompting & Context
Build better task briefs, supply evidence, use examples and structure output. Less incantation, more context engineering.
AI Safety
Privacy, prompt injection, deepfakes, connected tools, AI-written code and habits that reduce real-world risk.
AI Ethics
Fairness, accountability, meaningful oversight, explainability and data minimization without reducing ethics to slogans.
Practical Principles
Ten durable rules for using AI responsibly in a team, product or automated workflow.
Use the checklist →Standards & Regulation
EU AI Act, NIST AI RMF, UNESCO, C2PA and U.S. copyright guidance — with links to primary sources.
See the landscape →AI & Environment
Energy, data centres, water and why simplistic “one prompt equals X” claims can create false precision.
Read the evidence →Three rules that prevent a surprising amount of trouble
Give context, not magic words.
State the goal, audience, constraints, evidence and output format. Good inputs beat prompt folklore.
Ground important answers.
For facts that matter, provide or retrieve trusted sources and check that the evidence actually supports the conclusion.
Keep humans at irreversible gates.
Let agents draft and propose. Keep approval before payments, publishing, deletion, permissions and other high-impact actions.
One small practice, refreshed daily
Loading today's AI habit…