Getting started
What can you actually ask ChatGPT to do?
Practical ChatGPT requests for email, summaries, comparisons, and explanations, with a five-point check before you use the result.
4 min readRead the guide →
Choose the guide for your next task
Start with a concrete task. Each guide shows how to inspect the result and where the method has limits.
Everyday AI
Prompts, privacy, email, documents and practical comparisons.
Documents
Summarize a document with AI without missing the important parts
Use a coverage map, source references, and a separate register for amounts and dates before trusting an AI document summary.
4 min readRead the guide →
Practical decisions
Compare quotes and offers with AI without losing the numbers
A worked comparison of two service quotes: normalize the term, calculate total cost, surface missing conditions, and keep a source record.
4 min readRead the guide →
Learn AI
Three OpenAI Academy courses: a practical route
A beginner-friendly route through AI Foundations, Applied AI Foundations, and Agents and Workflows, with a task to try after each.
3 min readRead the guide →
Prompt basics
A simple ChatGPT prompt that leaves less to guess
A four-part prompt for beginners: goal, context, output format, and missing questions, tested against a vague request.
3 min readRead the guide →
Verification
How to check whether an AI answer is reliable
A practical check for AI-generated answers: separate facts from advice, trace primary sources, verify dates and numbers, and scale the effort to the risk.
4 min readRead the guide →
Privacy basics
What not to paste into ChatGPT
A simple checklist for passwords, client data, contracts, medical records, screenshots, and safer ways to ask for help.
3 min readRead the guide →
Everyday AI
Write clearer emails with ChatGPT without sounding robotic
A practical email-writing workflow: give the right facts, request two tones, remove boilerplate, and check the final draft before sending.
3 min readRead the guide →
Coding agents and development
Repository instructions, Git, tool permissions, review and security.
Security / Code scanning
GitHub Agentic Autofix: review the patch
A review protocol for GitHub Agentic Autofix: what CodeQL validation proves, how to test a proposed fix independently, and when to hold a draft PR.
9 min readRead the guide →
Agent security
Agent data injection: protect the trust boundary
A practical trust map for web and coding agents: external data, verified metadata, least privilege, approval gates and provenance.
4 min readRead the guide →
Agent tooling
Agent Plugins 1.0: skill, MCP and plugin explained
How a skill, an MCP server and an Agent Plugin differ, what portability covers, and what to inspect before installing one.
4 min readRead the guide →
ChatGPT Work
Start a ChatGPT task when a GitHub PR changes
A read-only pilot for GitHub pull-request event tasks: narrow filters, untrusted comments, minimum access and human review.
3 min readRead the guide →
Codex workflow
Codex in ChatGPT: a project safety checklist
Check the repository, scope, worktree, permissions, diff, tests and human review before accepting a Codex change.
3 min readRead the guide →
IDE agents
Set up Codex as a JetBrains agent provider
A first-session checklist for Codex via GitHub Copilot in JetBrains, with clear distinctions between provider, model, MCP, skills and plugins.
3 min readRead the guide →
Security review
Use Codex Security to scan a repository
A cautious first scan: inspect report, findings, manifest and coverage, validate a candidate issue, then compare a tested fix.
3 min readRead the guide →
Pull requests
Copilot PR approvals: keep a human merge gate
How to pilot Copilot approval assessments on low-risk paths without treating an AI approval as an independent human review.
3 min readRead the guide →
Agent sandbox
Test GitHub Copilot local sandbox boundaries
A low-risk pilot for Copilot local sandboxing: check filesystem and network boundaries, credentials and bypass approvals.
3 min readRead the guide →
MCP / GitHub Projects
Make Codex read the real GitHub Project card
A real source-confusion case and a fail-closed method to verify that Codex uses a GitHub Projects MCP tool instead of a related issue.
4 min readRead the guide →
Agent API
Pilot the OpenAI Agents API in a sandbox
A small observable pilot for a managed agent session: synthetic files, restricted network, isolated credentials, logs and success checks.
3 min readRead the guide →
VS Code agents
VS Code Agent Merge: pilot before auto-merge
A controlled rollout of Agent Merge for PR feedback, failed checks and conflicts, with a human gate for consequential merges.
3 min readRead the guide →
Agent workflow
Run parallel agent sessions in VS Code safely
Assign one owner per file, isolate writes and integrate sequentially when working with multiple VS Code agent sessions.
3 min readRead the guide →
Quick prompts
Three prompts to adapt
Use them on non-sensitive material and verify important claims.
A clear first request
prompt
Goal: [what you want to achieve]
Context: [only the details needed]
Format: [how the answer should look]
If something essential is missing, ask up to three questions before answering.Check an answer
prompt
Review this answer critically. Separate facts, inferences and suggestions. Identify dates and numbers I must verify, name primary sources to consult, and state what you are uncertain about.Compare offers
prompt
Compare these two offers without deciding for me. Show total cost over the same period, one-time fees, exclusions, risks, missing data and questions to ask. Show the formula for each total.Keep control
AI should help you think
It should not make a consequential decision for you without checks.
- Remove unnecessary personal data before pasting text.
- Verify numbers, dates, prices and conditions.
- Ask for assumptions and uncertainty.
- Use primary sources or a qualified person for high-stakes decisions.