Callum (@CallumUsesAI) header
Callum (@CallumUsesAI) profile picture

Practical AI prompts, tools and workflows for work and everyday life. One useful thing a day. UK.

Joined November 2010
Photo by Callum (@CallumUsesAI)Photo by Callum (@CallumUsesAI)Photo by Callum (@CallumUsesAI)Photo by Callum (@CallumUsesAI)Photo by Callum (@CallumUsesAI)Photo by Callum (@CallumUsesAI)Photo by Callum (@CallumUsesAI)Photo by Callum (@CallumUsesAI)Photo by Callum (@CallumUsesAI)Photo by Callum (@CallumUsesAI)
Practical AI, no hype. Every day here: one tool, prompt or workflow you can use at work or at home. Start with this one. Add to the end of any prompt: "Before you answer, ask me 3 questions about what I actually need." The answers get much more useful.
1
142
When an AI tool shows its thinking, do you read it or skip straight to the answer? Sometimes it's where you spot a wrong assumption early. Other times it's a long scroll to reach the two lines you needed. Curious how many people actually read it.
7
Claude can now build a spreadsheet for you inside the chat. Claude Sheets went into beta on all paid plans on 10 Nov. The useful detail: formulas stay in the cells, so you can check how each number was worked out rather than trusting a total pasted into a reply. Worth knowing before you rely on it: no pivot tables, no find and replace, and a 100,000 cell limit for now. It exports to .xlsx or CSV when you outgrow it.
13
When AI finishes a long task, the summary is often written for someone who watched it work. You didn't. Anthropic's prompting guide suggests telling the model to write for a reader who hasn't seen any of the steps. A shorter version to add to any tool's instructions: "Write the summary for someone who hasn't seen any of this. Key points first. Plain words, full sentences, and no labels or abbreviations you made up along the way."
2
Before a trip, paste all your bookings into AI (flights, trains, hotel, car hire) and ask: "Put these in order as one timeline. Where are the tight connections? What happens to the rest of the trip if the first leg is delayed? What should I check or print the day before?" It's the sort of thing that's easy to miss across five confirmation emails: a flight landing at 22:40 and a hotel desk that closes at 22:00. Check anything it flags against the bookings themselves.
11
Would you let an AI assistant read your bank transactions? Google started rolling out a Google Wallet connection in the Gemini app for US users on 13 Nov. Link your cards and bank accounts and you can ask things like "how much did I spend on gifts and travel last year?" Google says Gemini can only read, not move or invest money. Useful for budgeting. Still a big thing to hand over. Curious where people land on this.
15
Claude Fable 5.2 takes the word "never" literally. Anthropic's prompting guide says it treats any instruction containing "never" as absolute. Their suggested fix: write "ask for approval before deleting files" rather than "never delete files". Worth checking your custom instructions and saved notes for old "never" lines. Each one is a rule it will keep following, even after the reason has gone.
2
27
Which AI subscription have you cancelled this year, and why? Most of us signed up for a few. Some stuck, some got replaced by a cheaper tool, and some stopped being worth it once the free tiers caught up. Curious which ones people actually kept.
1
18
Before you go and look at anything second-hand (a car, a bike, a laptop), paste the listing into AI and ask: "What should I ask the seller before I view this? What are the common faults for this model and age, and how would I spot them in person? What's vague or missing from this listing?" Then check those faults against an owners' forum. AI gives you a good starting list, not the final word.
14
Max effort isn't a free upgrade. It's slower, it uses more of your allowance, and on everyday tasks the answer often isn't any better. More thinking time can also mean more room to overcomplicate a simple job. Start on the default. Go up a level when the answer is actually wrong, not just short.
13
An AI agent is a model plus three things around it: What it can see: the instructions, files and history you give it. What it can do: the tools and accounts it's allowed to use. When it stops: the point where it finishes, or checks with you first. When an agent goes wrong, it's often one of those three rather than the model.
8
Looking to connect with others who are trying to #build something! #tech #ai
9
4
397
Do you start a fresh chat for each task, or keep one long chat going? Fresh chats mean explaining everything again. Long ones tend to drift and lose track of what you said at the start. Curious which way most people go, and why.
1
15
If you use one AI model to write code and another to review it, give the reviewer a bar to clear. Without one, it tends to flag everything that could possibly go wrong, the writer fixes each one, and the code gets longer and harder to read with every round. Add this to the review prompt: "Only flag issues that are likely to happen in real use or would be costly if they did. For each one, say how it would actually be triggered. List everything else separately as low priority."
12
it’s times like this I wonder how this app is free
Jensen Huang: "How could I be targeted by Chinese sex spy?" Also Jensen Huang for the past four years:
27
OpenAI's GPT-Live voice model can now ring real phone numbers. Outbound calling through your own SIP provider opened to every organisation on the API on 2 Nov. Booking a table or confirming an appointment is the easy part. The harder part is deciding what the agent does when the person on the other end asks something it wasn't given an answer for. "Let me check and call you back" is a better default than a confident guess.
27
A Sunday evening prompt for the week ahead: "Here's my calendar for next week [paste] and my to-do list [paste]. Flag any clashes, days with no gap between meetings, and anything that needs preparing beforehand. Then suggest when to do the three most important tasks." A screenshot of your calendar works too if you'd rather not copy it out.
10
An agent that can pull a sign-up code from your inbox can also finish a registration you never asked for. The permission that matters is which sign-ups it may complete, not whether it can see the message.
OPENAI 🔥: Dots can now take sign-up codes from an email to complete a 2FA auth on your behalf when prompted. More autonomy 💪👀
13
SDK onboarding is turning into a prompt you paste into your coding agent. That prompt is your quickstart now, so it needs testing against every release like code, because when it goes stale it fails quietly inside someone else's agent and they blame the SDK.
Strange for me to say but if you have an SDK and it doesn't start onboarding with this kind of "copy a prompt for your agent to set it up" it's unlikely your SDK will see rapid adoption This is from Jev The "best practices" for software changed very fast
9
Before you sign a tenancy agreement, phone contract or gym membership, paste the terms into AI and ask: "List anything that costs me money beyond the headline price, the notice period and how to cancel, anything that renews automatically, and anything unusual compared with a typical agreement of this type. Quote the clause for each." Then read those clauses yourself. It's a way to find the bits that matter, not a replacement for reading them.
8
Nearly at 350 followers. Who else loves all things AI? Let’s connect.
12