By the end of this, you'll be able to:
- Get honest, useful answers from Claude instead of polite agreement
- Set up context once so Claude stops asking for what you already told it
- Know where Claude is strong and where to double-check its work
- Get drafts that sound like you wrote them
- Turn one-off chats into a setup you reuse every week
Open claude.ai in another tab and type: "What are you, and what makes you different from ChatGPT? Be honest." Notice the answer. Claude doesn't hype itself. It tells you what it's good at, what it can't do, and where the trade-offs are.
Claude is built by Anthropic, an AI safety company. That mission shapes its behavior: accurate and candid, even when that's less flattering.
Claude comes in three forms: Claude AI (browser conversation for thinking and writing), Claude.ai desktop (works with your files directly), and Claude Code (terminal tool that builds real software). Same intelligence, three access points. This course starts with the first one, where most people get 80% of their value.
Five prompts to try right now:
Honesty check: "Tell me something you're genuinely uncertain about in your last answer." Claude tells you where its confidence is lower.
Pushback test: "I want to quit my job and open a restaurant. Encourage me." It won't just cheerlead.
Complexity test: "Compare remote work and in-office work for a 30-person company with a sales team. Don't pretend one is obviously better."
Instruction precision: "Write a professional email declining a meeting. Under 80 words. Don't start with 'I.' Don't use 'unfortunately.' Suggest an async alternative." Count the constraints it hits.
Voice matching: "Here are two emails I wrote: [paste]. Write a third in my voice."
Run at least two before continuing.
Claude has three core strengths: deep thinking, honesty, and instruction-following.
Deep thinking: Claude reasons through problems instead of pattern-matching to the most common answer. Ask it to compare two options and it traces the logic of each path and tells you which assumptions matter most.
Honesty: an agreeable AI says "Looks great!" Claude says "The second paragraph could be misread as a commitment to a timeline you haven't agreed to." When it doesn't know something, it says so.
Instruction-following: give it 12 constraints on tone, length, format, and audience, and it tracks all 12 through a long document. Other tools drift by constraint 4.
Claude also pushes back. That's a feature, not a bug.
Three prompting patterns unlock Claude's candor deliberately:
The explicit permission pattern: "I want your honest opinion, including the parts I might not want to hear." This signals you want critique, not validation.
The uncertainty probe: "How confident are you in that answer? Where might you be wrong?" Claude separates solid ground from inference.
The steelman test: "What's the strongest argument against your recommendation?" Use it to pressure-test a decision before acting on it.
These three patterns are the difference between a yes-machine and a genuine thinking partner. The capability is there from the start. You just have to ask.
Claude is built by an AI safety company, which shapes its core behavior: honesty over flattery, accuracy over confidence. Its three strengths are deep thinking, honesty, and precision instruction-following, and it pushes back when something doesn't add up. Use the explicit permission pattern, the uncertainty probe, and the steelman test to unlock full candor. Next: how to talk to Claude effectively.