Pinned Tweet
Cooking something new for @X and on X.
We all can build apps but #distribution is the challenge. Let's #fix this one and for all.
1
366
Abhishek | Klio + Vex retweeted
$klio is apparently being impersonated and is a fake
@7hakurg is the original github owner and @klio_tech is leeching off it
still a little above entry, protect your capital
still holding my $klio after a first run up
product seems good
ath is still attainable and good ROI
0x59d80b8d4593363363f2ff4232cbc8883fd429db
Abhishek | Klio + Vex retweeted
I was losing an hour in CapCut on every reel - cutting dead air, captioning every word, reframing to 9:16.
So I built LoopReel.
Drop in any raw video, AI does the whole edit, you get a ready-to-post reel in ~2 min.
Shipping AI agents without guardrails is a trap I fell into.
For months, I measured success by "did it run?" Not by output quality. Not by drift over time. Not by cost per task.
That's why most agents quietly fail in production.
I'm fixing that. Building in public.
#AIAgents
3
227
I left a stable dev job to build something that actually solves agent memory.
I have one idea.
I have 90 days.
I have to ship.
Documenting everything from day one.
#BuildInPublic
2
191
Persistent memory isn't a nice-to-have anymore. It's the baseline.
When your agent forgets context mid-task, you don't have an agent.
You have an expensive autocomplete.
Stop patching stateless loops. Build for memory from day one.
#AIAgents
78
Most AI agents fail in production because nobody defined what "good enough" looks like before shipping.
Here's how we think about it:
1. Define your failure modes first. Not edge cases. The actual ways your agent can silently go wrong. Hallucination, drift, low confidence outputs, blocked actions.
2. Set a threshold before you deploy. What confidence score triggers a human review? What output pattern gets flagged automatically? Write it down before your first real user hits it.
3. Log everything from day one. Not just errors. Every input, every decision, every output. You can't debug drift you didn't capture.
4. Build a feedback loop into the product itself. Not a quarterly eval. Something that surfaces degradation in real time so you can act before users notice.
5. Treat your verification layer like infrastructure. Not an afterthought. It belongs in your architecture the same way logging and auth do.
Most teams skip all of this until something breaks. Then they're debugging blind with production data and frustrated users.
Define what failure looks like before it finds you.
53
The most dangerous assumption is thinking you need a finished product to start building in public.
Ship the idea. Get the feedback. Then build what people actually want.
Building in the open from day one.
#BuildInPublic
40
🚀 Want to build a #Resend-like product (dev-friendly transactional email API with templates, analytics, webhooks) but 5-10x cheaper?
Skip Resend's markup + #Amazon SES costs. Use open-source @useSend_com + ZeptoMail. Full control, no vendor lock-in. Thread 👇
2
71
Why this beats Resend/SES:
- 80-90% cheaper at scale
- Full ownership (self-hosted)
- Same dev DX
- Better for transactional-only (ZeptoMail's specialty)
- No surprises on bill
Already seeing people self-host UseSend on SES → this is the smarter evolution.
1
17
Ready to ship your own email infra?
Star the @useSend_com repo, try ZeptoMail's free credits, and fork it today.
Questions? Drop them below - happy to share code snippets or Docker compose.
What would YOU build with this? 👇
#BuildInPublic #DevTools #EmailAPI #OpenSource
1
27




