I’ve been experimenting with voice typing for AI prompts because long prompts are often easier for me to speak than type.
I came across a video where someone built their own dictation tool, which led me to the typr GitHub repo.
Note: this is public-source / license-pending while I clarify upstream licensing.
I’ve credited the original repo and opened an upstream license clarification request.
I’ve been experimenting with voice typing for AI prompts because long prompts are often easier for me to speak than type.
I came across a video where someone built their own dictation tool, which led me to the typr GitHub repo.
The most meaningful customization for me was Urdu-specific support, because multilingual workflows are where generic tools often fall short.
Repo:
https://t.co/fUtR90XOzM
Release:
https://t.co/CmHQoV5a6J
A lot of founders do not need a better CRM first.
They need fewer dropped leads.
Simple AI workflow:
- capture every inquiry
- summarize the prospect's problem
- classify fit and urgency
- draft the reply
- remind the founder if nothing was sent
The win is not sophistication. It is fewer missed conversations.
Agree on the direction. I would still keep the first version boring: automated onboarding should collect context, classify the user path, draft the next step, and escalate weird cases. The agent is leverage only if the fallback is obvious; otherwise it becomes another support job.
Most founders do not need "more AI tools".
They need 2-3 boring automations that remove repeat work from the business:
- qualify inbound leads
- summarize customer calls
- draft follow-ups
- monitor competitors
- turn notes into content
- route support requests
- update internal docs
I am building practical AI workflows for founders and solopreneurs.
For the next 30 days I am sharing:
1. bad AI automation advice to ignore
2. workflow teardowns for real businesses
3. small automations built in public
4. practical ways to save 5-10 hours/week
If you are drowning in repeat ops, DM "workflow" and I will suggest one automation for your business.
My flows ran for hours on Opus 4.6 — now on Opus 4.7 they’re dead in 30 minutes. New tokenizer and deeper thinking are absolutely eating tokens alive. 5-hour limit? Gone in half an hour. Who else is getting destroyed by this? #ClaudeAI
Mid-size engineering teams in emerging markets are at a quiet crossroads in 2026.
Most are still bolting AI onto old processes. The ones pulling ahead are redesigning workflows around agentic AI — where humans direct and agents execute the full SDLC.
From my Director Chair: the real bottleneck isn’t the technology. It’s building trust and shared context.
What’s the biggest blocker stopping your team from handing real work to agents right now?
The way engineering teams are using AI has changed more in the last 6 months than the previous 6 years.
I've been exploring different frameworks, Compound Engineering, Spec-Driven Development, AI-DLC — and what strikes me is that everyone is solving the same problem differently:
Curious what's actually working at your org:
→ How are you managing institutional knowledge with AI tools?
→ Are your devs building shared memory systems or is it still per-person?
→ What's the one thing you wish you'd set up earlier?
Started with Antigravity, ended up using the latest Codex app, and learned a lot about new AI workflows by shipping something real.
More and more, I think customized apps are the future with AI.
GitHub: https://t.co/fjbzlwM9Dt: https://t.co/ERqXEkSUoc
Needed to merge a few PDFs this weekend, and instead of searching for a tool, I built one.
That turned into KagazKit, a small open-source desktop app for PDF tasks.
I'm loving my Web3 learning journey on Metaschool! 🎉 Free Web3 courses in a gamified learning setup—Quests, XPs, and even NFTs when you finish! 🤩 Let's do this together! https://t.co/XeLmcO3OeB