Instagram story viewer> @toyafounder> Posts

Anishek Kamal | AI engineer

@toyafounder

139

followers

187

following

AI engineer at Microsoft · builder
Running an AI-built IG experiment in public
I use AI to research, write & design the carousels.

POSTS STORIES REELS TAGGED

@toyafounder is a public experiment: I use AI to research, write, and design the carousels — and I'm testing how far an AI-built niche Instagram can scale.

Weekdays you get stealable skills and roadmaps. Twice a month I'll post a real update on what's working.

Save this brief. Share it. Follow the test.

AI Instagram experiment · AI carousels · Anishek Kamal · AI engineer · Save share growth by @toyafounder

6

0

a month ago

Hey — I'm Anishek Kamal. I work at Microsoft as a Senior Data & AI Architect, and I'm a builder who ships small tools and products for my own problems (and hopefully yours). I also advise AI and health-tech startups.

@toyafounder is my public AI-built Instagram experiment: AI helps research, write, and design the carousels. I'm testing how far that can scale.

Steal the stacks. Follow the experiment.

Anishek Kamal · Microsoft · AI engineer · builder · startup advisor · AI carousels by @toyafounder

40

0

a month ago

Using Opus for everything? That’s where your Claude limit goes.

Anthropic’s own guide puts it plainly: Sonnet uses your usage at a moderate pace, Opus is heavy. And yesterday (Sep 28) Anthropic released Claude Sonnet 5.5, on every plan including Free.

What changes for you:
1. Tap the model name by the message box and pick Sonnet 5.5 for everyday work: emails, summaries, docs, slides, spreadsheets
2. Stuck? Turn up Effort in the same menu before you jump to Opus
3. Save Opus 5.5 for open ended work: strategy, long research, big decisions

We ran the same client email through both. Both were fine to send. Sonnet’s was shorter.

Only use claude.ai or the official Claude apps. Ignore “try Sonnet 5.5 free” lookalike sites.

Send this to the friend who runs out of Claude by lunch.

#claude #anthropic #aitools #productivity taken in San Francisco, California by @toyafounder

11

1

14 days ago

3

0

19 days ago

Your friends love it. ChatGPT says it’s a great idea. Your mom says quit your job.

None of them will pay you.

Test it in one day instead:
1. Find the one customer who hurts most
2. Find real complaints (web search on, open the links)
3. Set kill criteria before you build
4. Talk to 5 real people, no leading questions

We ran “is this a good idea?” vs the skeptic prompt. 447 polite words vs 3 numbers you can act on today. Slide 8.

Part 1 of 3 in this week’s founder series. Friday: what to write before you build. Monday: where to launch.

Send this to the friend with a new startup idea every weekend.

#startup #founder #chatgpt #claudeai #startupideas taken in California USA by @toyafounder

8

0

19 days ago

5 minutes, one switch per app:

ChatGPT: Settings, Data controls, “Improve the model for everyone”
Claude: Settings, Privacy, “Help Improve our AI models”
Gemini: Settings and help, Activity, Keep Activity

For one-off sensitive chats: ChatGPT Temporary chat or Claude Incognito.

And no setting makes it safe to paste passwords or client data you aren’t allowed to share.

Send this to the friend who pastes work docs into ChatGPT.

#aiprivacy #chatgpt #claudeai #gemini by @toyafounder

3

0

20 days ago

On Sep 14, OpenAI retired automatic switching from Instant to Thinking for Plus and Pro. If ChatGPT feels less careful, that might be why. You pick the mode now.

Rule of thumb: if you could check the answer in 10 seconds, go fast. If a wrong answer costs you, make it think.

Instant, Thinking, Deep research, Study mode and Temporary chat, and when to use each, are in the slides.

#chatgpt #openai #aitips #productivity by @toyafounder

3

1

21 days ago

AI sounds like AI because you gave it a topic and no examples.

The fix takes 15 minutes once:

1. Collect 3 to 5 things you actually wrote
2. Make it describe your style as rules, with quotes as proof
3. Fix the rules and save them in custom instructions or a Project
4. Give it a job with an audience, not a topic
5. Last pass: “mark every sentence I wouldn’t say out loud”

The prompts are in the slides. Save this before your next draft.

Want 5 more copy-paste prompts for your workday? Comment WORKFLOW and I’ll DM them to you.

#aiwriting #chatgpt #claudeai #contentcreation by @toyafounder

3

0

22 days ago

All on free plans:

1. ChatGPT Study mode (type @study)
2. ChatGPT quizzes
3. ChatGPT scheduled tasks (3 free)
4. Gemini Guided Learning
5. Gemini Notebook’s Critique audio overview (NotebookLM got renamed)
6. Claude Incognito chat

Save it and try one tonight.

#aitools #chatgpt #gemini #claudeai by @toyafounder

3

2

23 days ago

OpenAI, Anthropic and Google all publish how to prompt their models. 7 rules from those guides, with who said each one:

1. Paste first, ask last (Anthropic: up to 30% better in their tests)
2. Say what to do, not what to avoid (Anthropic)
3. Tell it why (Anthropic)
4. Show 3 to 5 examples (Anthropic)
5. Describe the result, not every step (OpenAI)
6. Drop “be concise” (OpenAI)
7. Persona, task, context, format (Google)

Save it for your next prompt.

#promptengineering #chatgpt #claudeai #gemini by @toyafounder

3

0

24 days ago

Jev doesn’t write answers. It makes picks.

TypeSafe AI, a startup led by an ex-OpenAI researcher, built it for the small decisions software makes all day: which queue your ticket goes to, which email gets flagged, whether an answer is good enough to send.

You get a label and how sure it is. That’s all.

Steal the idea today: when you want a decision from ChatGPT or Claude, give it the options and ask for one pick and a confidence from 0 to 100. The prompt is in slide 7.

One warning: lots of “try Jev free” sites showed up this week. The company’s own site is typesafe.ai.

Send this to the friend who thinks AI means chatbot.

#ai #aitools #chatgpt #artificialintelligence by @toyafounder

3

1

24 days ago

Five words in, generic answer out, then “AI is overrated.”

The model usually isn’t the problem. The input is. 6 swaps:

1. Brief it, don’t search it
2. Ask for options and trade-offs
3. Make it critique its own answer
4. Ask how your plan fails
5. Paste first, then give it a job
6. Keep your context in a Project

Works with ChatGPT, Claude and Gemini. Which one do you already do?

#chatgpt #claudeai #aitips #promptengineering by @toyafounder

3

1

25 days ago