Practical AI tools for everyday life and business. Explore what AI can do
Learning Hub
PRACTICAL AI FIELD GUIDE

Claude & Claude Code workflows

Start by understanding the project, then make changes with a clear way to check the result.

Free guide12 min readReviewed

Claude chat and Claude Code are different working environments. The code tips require Claude Code with access to your project. We reviewed the linked talk’s auto-generated transcript and cross-checked product details against current Anthropic docs.

Your first useful task

Start with one small win

  1. Open a backed-up project and ask one question about how it works, without requesting edits.
  2. Choose one small change. Explain the symptom and what success would look like.
  3. Provide the project’s real test command or a visual reference, then inspect the result and changes.

Practical tips that make a difference

Set success criteria before polishing a prompt

For writing or analysis in Claude chat, decide what a good answer must contain and how you will assess it. If the output misses the mark, change one instruction or example and compare the result. Prompt wording is not a substitute for a usable evaluation.

Based on: Anthropic · Prompt engineering overview ↗

Supply what the repository cannot reveal

Name the user-visible symptom, the preferred behavior and constraints such as “keep the current database structure.” Claude Code can inspect code patterns, but it cannot infer an unpublished business decision. Add the missing decision rather than a large generic instruction manual.

Based on: Anthropic · Claude Code best practices ↗

Make verification concrete

Give the agent a relevant test, build command or screenshot comparison. Ask for the observed result, not an assurance that things “should work.” A successful build checks one part of the system; it does not prove account access or a payment flow works.

Based on: Anthropic · Claude Code best practices ↗

Creator’s workflow

Boris Cherny’s practical talk: lessons worth using

Timestamped paraphrases from the linked recording, uploaded by Solomon Alfred. Auto-captions sometimes misname Claude; we use the correct product name. These are historical workflow lessons, not a current feature or privacy guarantee.

4:00 ↗

Start with questions

Learn the codebase before asking for edits.

5:49 ↗

Ask about history

Use Git history to understand a surprising design decision.

8:22 ↗

Plan larger changes

Agree on an approach before a substantial implementation.

11:02 ↗

Close the feedback loop

Provide tests or screenshots the agent can check.

12:22 ↗

Keep shared context concise

Store useful project commands and conventions in CLAUDE.md.

22:23 ↗

Isolate parallel work

Separate checkouts or worktrees avoid overlapping edits.

25:02 ↗

Use visual references

Give the agent a mockup instead of describing every pixel.

Watch Boris Cherny’s talk on YouTube ↗

Developer notes

Keep CLAUDE.md focused

Record the project-specific commands and gotchas that save repeated rediscovery. Keep long procedures separate. Natural-language instructions guide the model; hard access boundaries require permissions, sandbox controls or enforced checks.

Based on: Anthropic · Project memory and instructions ↗

Use today’s shortcuts

The current terminal reference lists Ctrl+R for searching prompt history and Ctrl+O for the detailed transcript. The linked talk describes an older Ctrl+R behavior. Use the current reference for your terminal and version rather than memorising every shortcut from a recording.

Based on: Anthropic · Claude Code interactive reference ↗

Give feedback without starting over

Use a focused follow-up to preserve the good parts and correct the failing part. For a design task, provide the actual screenshot and ask for a comparison after the change. This is our recommended review pattern, not a guarantee that an agent will catch every defect.

Based on: Anthropic · Common coding workflows ↗

Copy, adapt, then check

Original Utilising AI examples. Replace bracketed placeholders before use. These buttons copy text only; they do not run code or contact an AI provider.

First conversation with an unfamiliar codebase

I am new to this project. Explain how [feature] works, where the main logic lives, and which tests cover it. Point to files and explain unfamiliar terms. Read-only for now; do not change files.

Original example inspired by the talk’s codebase Q&A approach.

Based on: Boris Cherny talk · Practical Claude Code tips ↗

Plan a meaningful change

I want [specific behavior]. Current behavior: [example]. Constraints: [requirements]. Explore the relevant code and propose an approach with affected files, trade-offs and checks. Stop after the plan so I can review it.

Use the plan step when choices or scope need agreement. Tiny fixes may not need it.

Based on: Anthropic · Common coding workflows ↗

Project instructions starter (CLAUDE.md)

# Project notes
- Product behavior to preserve: [describe].
- Run one relevant test: [actual command].
- Typecheck/build: [actual commands].
- Project gotcha: [one important exception].
- Definition of done: [observable acceptance criteria].

A template to adapt, not a runnable command. Confirm all commands exist. Instruction files are not security enforcement.

Based on: Anthropic · Project memory and instructions ↗

Check your result

  • Is the correct project open and its starting state backed up?
  • Can you see test output or a before/after screenshot?
  • Have you reviewed the diff and any changed dependency or permission?

Common mistakes to avoid

  • Copying old installation or shortcut advice without checking current docs.
  • Treating the talk’s historical privacy statements as today’s data-handling policy.
  • Granting broad command or connector access just to avoid prompts.

Sources & review notes

Reviewed on 15 November 2026. Tips are paraphrased from the linked documentation; examples and practical recommendations are our own. This is a documentation review, not a hands-on test of every provider account or feature. Interfaces, plans and commands can change.

Independent learning resource by Utilising AI Pty Ltd. No endorsement by the named providers is claimed. Review your chosen tool’s permissions and data handling before supplying business information.

Keep building your AI skills

Create your free assistant prompt Browse all six guides
Claude & Claude Code workflows | UtilisingAI