Today's top AI questions, answered
1How do I stop an AI coding assistant from leaking passwords into my code or public repos?
Assume the AI should never see a secret. Keep credentials in environment variables or a secrets manager, never in source files, and add .env files to .gitignore.
Be careful with "always allow" or "don't ask again" options. Saved permission rules can end up containing pasted keys or full commands, and they are often stored in project files that get committed. Prefer narrow, specific approvals over blanket ones, and review any settings files before you commit them. Add a secret scanner as a pre-commit hook and in your CI pipeline. If a key does leak, rotate it immediately. Deleting the file or commit is not enough, because copies may already exist. Check your tool's current docs for where permissions are stored.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
2How do I choose between closed and open AI models when both cost and quality matter?
Start from your task, not a leaderboard. Collect 20 to 50 real examples from your work, run them through two or three models, and score the outputs yourself.
Often a smaller, cheaper model handles routine jobs like summaries, tagging and first drafts. Save the strongest model for hard reasoning or high-stakes output. Open models make sense when data must stay inside your environment, when you have high volume and want predictable costs, or when you need to customise. Count the hosting and maintenance effort too. Closed models are easier to start with and usually need no setup. Prices and rankings change often, so recheck current pricing pages and rerun your test set every few months.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
3Is it healthy to use ChatGPT as a companion, or should it stay a work tool?
It can be both, as long as you stay clear about what it is. As a thinking partner it is useful. You can talk through a decision, rehearse a hard conversation, or ask it to challenge your plan.
It does not know you the way a colleague or friend does, it tends to agree with you, and it can sound confident while being wrong. A few guardrails help. Ask it directly to disagree or list the weak points in your thinking. Avoid sharing confidential or sensitive personal details. Keep real people involved in decisions about health, money and relationships. If you notice you are using it to avoid human conversations, step back. Review the privacy and memory settings in your account.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
4How do I write a prompt that makes an AI work like a careful researcher?
A process works better than a job title. State the question, why it matters, who the audience is, the scope, and which kinds of sources are acceptable. Then specify the output format, such as a one-page brief with a short summary first.
Ask it to plan first by listing sub-questions, then gather evidence, then synthesise. Tell it to separate facts, inferences and unknowns, and to say "not found" instead of guessing. Use a tool with web search, or paste in your own documents, so it works from real material. Ask for links or page references. Open the key sources and check them yourself, because AI can invent citations that look convincing.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
5How do I stop a voice AI agent on a phone call from hearing and answering its own voice?
This is an echo problem. The agent's audio plays through the speaker, the microphone picks it up, and the model treats it as the caller speaking.
Try these fixes in order. First, turn on echo cancellation and noise suppression in your audio capture or telephony layer. Second, lower the speaker volume or use a headset when testing. Third, mute or ignore the incoming audio while the agent is speaking. This is simple and reliable, but it stops callers from interrupting. Fourth, tune turn detection by raising the voice activity threshold and allowing slightly longer silence before a turn ends. Always test on a real speakerphone, since laptop tests hide the issue. Check the realtime API docs for current turn detection options.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
Earlier questions
7 November 2026
1How can I spot AI deepfake scam ads and protect myself and my team?
Treat any ad with a famous face, an urgent deadline or a promise of guaranteed returns as suspect. Look for tells such as lips slightly out of sync, odd blinking, a flat voice, or blurry edges around the face and hands. Do not rely on these alone, because good fakes are hard to catch.
Verify instead of judging by looks. Search for the claim on the person's official website or verified channels, inspect the web address before clicking, and never move an investment conversation to WhatsApp or Telegram. For teams, set a rule that any payment or password request needs a second confirmation through a phone number you already know. Report the ad on the platform and warn colleagues, since scammers reuse the same creative. If money has been sent, contact your bank and the national cybercrime reporting portal immediately.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
2When is a small, fast AI model good enough, and when should I use a bigger one?
Fast, lightweight models suit clear, low-risk tasks: summarising a meeting, sorting emails, reformatting data, drafting first versions, or pulling fields out of documents. They respond quickly and usually cost less, which matters when you repeat a task hundreds of times.
Move to a larger model when the task needs multi-step reasoning, careful reading of long or messy documents, nuanced writing, or when a wrong answer is expensive, such as legal, financial or production code work. A simple test: run ten real examples of your task through both models and compare the results side by side. If the small one passes nine, use it and keep the bigger one for the exceptions. Model lineups change often, so check the vendor's current documentation.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
3How do I get an AI model to follow my writing style instructions every time?
Models drift from style rules when instructions are long, vague or buried. Put them in a short, separate block at the top, in the system prompt or custom instructions, and write concrete rules. "Sentences under 20 words, no bullet points, British spelling" works far better than "be friendly and professional".
Add one or two short sample paragraphs in the style you want, because examples beat descriptions. Also state what to avoid. In long chats the style fades, so repeat a one-line reminder or start a fresh chat with the same instruction block. For recurring work, save the block as a template, project or custom assistant. If output still drifts, ask the model to restate the rules before it drafts.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
4What should I do when an AI chat says its memory or context window is full?
Two different limits are involved. The context window is how much text the model can consider in one conversation, including your messages, its replies and pasted documents. Saved memory is a separate store of facts about you that carries across chats. When either fills up, quality drops or you see errors.
For a full context, ask the AI to summarise decisions, open questions and key facts, then paste that summary into a new chat. Avoid pasting whole documents repeatedly; share only the relevant sections. For full memory, open the settings, delete outdated items and merge duplicates. Keep one project per topic so each chat stays focused. Exact limits vary by tool and plan, so check the current docs.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
5How can I tell if heavy AI use is really making me more productive?
More prompts do not equal more output. The real measure is time to a finished, usable result, including your own checking and fixing. Pick three recurring tasks, such as a weekly report, a client email and a research summary. Time them for a week without AI, then for a week with AI, counting the full effort including corrections. Note how many fixes each draft needed and whether quality improved.
Typically AI helps most with first drafts, summaries and structuring, and least with tasks that need precise facts. If you keep correcting the same mistakes, improve the prompt with context, examples and a format, or stop using AI for that task. Review the log monthly and keep only the uses that pay off.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
6 November 2026
1How do I keep AI agent costs under control as my team uses them more?
Agents cost more than chatbots because they loop: they read files, call tools, retry and re-read. Spend grows with every step, not every prompt.
Set a hard spending cap and usage alerts in your provider dashboard before anything else. Use a cheaper, smaller model for routine steps and keep the strongest model for planning and hard problems. Keep context lean by giving narrow tasks, naming the exact files or documents, and starting a fresh session instead of letting one thread grow endlessly. Limit retries and step counts. Then track cost per finished task rather than per month. If an agent costs more than the human time it saves, redesign the workflow. Check your tool's current pricing and limits pages for details.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
2How can I stop personal details leaking when I use AI coding tools?
Assume an AI tool can see anything in the folder it works in, plus things like git settings, environment variables and any web access you allow. Treat that as a data-sharing decision.
Practical steps: read the tool's privacy and data-handling settings and switch off anything optional. Review its permission prompts instead of approving everything. Keep secrets, customer data and personal files out of project folders, and exclude .env files. Use a no-reply or work-only email in your git config. For sensitive work, run the tool inside a container or separate user account. Finally, check the vendor's current documentation on what is sent and stored, and ask your IT team to confirm the policy before rolling it out.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
3How do I let AI agents do real work without risking costly mistakes?
Start by limiting what an agent can touch. Give it read-only access first, and add write or send permissions only for specific, low-stakes tasks. Work in a copy, a draft folder or a test environment, not in live files.
Keep a human approval step before anything irreversible: sending email, deleting data, spending money, publishing. Ask the agent to explain its plan before it acts, and keep a log of what it did. Check results on a small batch before running the full job. Make sure you can undo changes, with backups or version history. Expect occasional confident errors, and build your process so one error cannot do serious damage.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
4How do I get an AI coding assistant to reuse shared code instead of duplicating it?
AI assistants tend to write fresh code for each task because they only see what is in their context. Fix that by making the shared code visible and by stating your rules.
Create a short project instructions file (many tools read one automatically) saying where shared utilities live, how they are named, and that new logic should reuse them. Point the assistant at the specific shared files in your prompt. Ask it to search the codebase for an existing function before writing a new one. For larger changes, ask for a plan first and check it for reuse. Afterwards, request a review pass that looks only for duplicated logic and proposes moving it into a shared module.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
5Why does ChatGPT forget details I already gave it, and how can I fix that?
Chat models only work within a limited context window. In long conversations, early details get dropped or deprioritised. New chats also start blank unless a memory feature carries things over, and that feature saves only selected facts.
To reduce forgetting, keep one chat per task and restate key facts when a thread gets long. Keep a short reusable brief with your role, audience, format and constraints, and paste it at the start. Use the custom instructions or project features for things that never change, and check the memory settings to see what is saved. For long documents, ask for a summary of the decisions so far and begin a new chat with that summary. Check the tool's current help pages for how memory works.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
5 November 2026
1How can I use AI to start clearing years of messy legacy code without breaking things?
Start with understanding, not rewriting. Ask the AI to explain a module, map what depends on it and list the behaviours it seems to have. Then have it write tests that pin down what the code does today, before you change anything. Those tests are your safety net.
Work in small slices, such as one function or file per change, with a human review and a test run each time. AI is good at dull jobs like documenting, renaming, upgrading libraries and removing dead code. Keep architecture and risk decisions with your team. Also write a short guide to your conventions for the AI to read. Well-documented code gives better AI output too.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
2How do I check whether the AI features in my work tools are sending company data elsewhere?
Treat every update as a possible change. After updating a work tool, open its settings and search for words like AI, assistant, plugin, telemetry and sharing. Turn off anything you have not approved.
Then read the vendor's data processing or privacy page to see what is sent, where it goes and whether it is stored. If you have an IT or security team, ask them to review it and set admin-level controls. For tools that handle API keys, customer records or internal endpoints, use test data instead of production data, and keep secrets out of shared collections and workspaces. Prefer tools that let admins disable AI features by default.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
3How do I decide which parts of my daily work to hand over to AI?
For one week, jot down the tasks you repeat. Then pick the ones that are text-heavy, repetitive and low-risk if slightly wrong, such as first drafts, meeting summaries, reformatting, email replies and early research.
Keep judgement calls, final facts, sensitive decisions and anything client-facing under your own review. Calibrate first by testing the AI on a task you know well, so you can see where it slips. Treat its output like work from a quick but careless intern: useful, but always checked. When a prompt works, save it as a template so the gain repeats. Review the list monthly and add one new task at a time.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
4What is a sensible starting AI toolkit for a professional, and when do I need more than one tool?
Start with one general assistant and use it daily until you know its strengths. Move to a paid plan only when you regularly hit limits or need features that matter to your work. Check each vendor's current plans, since they change often.
Add a second tool only to fill a clear gap, such as a different model for a second opinion, or a coding assistant if you write software. Pay-as-you-go APIs make sense once you automate repeated tasks. Local models are worth a look when data must stay in-house. Before adding anything, track for a month which tools you actually open. Most people need fewer tools than they think.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
5How do I stop my chatbot conversations from being used for training or read by human reviewers?
Check the privacy and activity settings in each tool. Look for options on chat history, activity and model improvement. Some tools only let you opt out by turning history off or by using temporary chats, so confirm what your plan allows.
Work or enterprise accounts often have different data terms from personal ones, so use the approved tool for company material. In a personal account, never paste client data, passwords, unreleased financials or personal identifiers. Replace names and numbers with placeholders before asking for help. Policies change, so reread the provider's current privacy page now and then, and ask your IT team if you are unsure.
Answered by Shantanu Anand, AI Trainer & Consultant
Want your team doing this hands-on? See Shantanu's AI training →
Want your team to use AI like this?
Hands-on AI training for professionals and teams, online or in person anywhere in India. Get the training details on WhatsApp, or book a free 20-minute call.
Get training detailsBook a free call