this is what happened during w2 of project shipyard s3 by @AILearnCircle
we had @Connect_Mani from @SarvamAI come in and talk about something every builder needs to hear:
"don’t just build. build with intent. build with good product sense. and yes talk to users a lot more often"
he spoke about product sense, how to think about what you’re building, what you should actually spend time on and what makes a product genuinely good.
Mani also shared some super interesting stories from his time at Sarvam + Sable Money. loved the whole session.
then Varun lash from Aspire took us into the GTM world
4 pillars, how to think about each one, what to experiment with and how he’s seen this play out through his journey at Zolve + Aspire.
big big thanks to @useflo101 , @avneer_bh for hosting us and being such a solid partner for shipyard s3 ❤️
we ended the 10-hour sprint with every builder sharing what they’ve done over the last 2 weeks.
also, @Abhidwivedimany casually ended the day with some beatboxing 😂
now the GTM arc begins.
HSR was fun.
Indiranagar, we’re coming.
super excited to see what these builders ship next.
to all the builders who are misfits but are creating awesome stuff @MohithSarmaKLK , @atypical_chai , @mahaulguy , @_vin_mehta_ , @AditiDewan12 , @joshibhai , @RahulRanjan_R, @notbhuvanesh, @ErKiranKumar_ , @renukarajpuria, @vishalphdk, @RaviVidharshana and many others
People, I have notesssss!
Speaker Background:
Mani, product leader at Sarvam (AI foundational models + application layer)
Journey: customer care → digital marketing/analytics → solopreneur → product management
Forced to learn coding as a solopreneur; built tools over weekends, sold to first customers
Subsequent stints: Stable Money (product + design), Sarvam (2.5 years)
Sarvam product team today: 20 PMs, zero designers, three frontend engineersAll design done directly by PMs using AI-native systems
Can AI agents agree?
Communication is one of the biggest challenges in multi-agent systems.
New research tests LLM-based agents on Byzantine consensus games, scenarios where agents must agree on a value even when some participants behave adversarially.
The main finding: valid agreement is unreliable even in fully benign settings, and degrades further as group size grows. Most failures come from convergence stalls and timeouts, not subtle value corruption.
Why does it matter?
Multi-agent systems are being deployed in high-stakes coordination tasks. This paper is an early signal that reliable consensus is not an emergent property you can assume. It needs to be designed explicitly.
Paper: https://t.co/3fllhchiKX
Learn to build effective AI agents in our academy: https://t.co/1e8RZKs4uX
@mwseibel I've developed two frameworks. One for cognitive mismatch and other one is archetype and blurred roles in this hyper connected world (leads to isolation too). Haven't really posted anything so far but seeing your post and these comments helped me see the urgency of it.
@mwseibel I'm an INTP. MBTI helped a bit but couldn't explain other traits. Later discovered ADHD (pi), dyslexia. "Disorder" didn't make sense when i looked around at autism and other "DEVELOPMENT DISORDERS". We were merely outliers in the system. Noticed cognitive friction is the issue.
@mwseibel Yes, mine was- i was curious, thought everyone is. thus grown ups must know a lot. Started seeing inconsistency as i grew older (12-13 yo). That led me to assume ggat everyone can see these patterns but choosing to be ignorant. Later MBTI helped me see diff cognition
This is the magic I've been waiting for, what AI can do:
"There is the video of a squirrel somewhere in the folder, convert it to mp4"
Anthropic did it. CUA will finally become useful - this is a historic moment.
⚡ Building enterprise agents at Coinbase with LangSmith ⚡
Coinbase went from zero to production AI agents in six weeks, then cut future build time from 12 weeks to under a week.
Their Enterprise AI Tiger Team built a "paved road" so any team could ship agents the same way they ship code.
What made this work:
→ Code-first graphs with LangGraph & LangChain over low-code tools. Typed interfaces and unit-testable nodes beat prompt engineering for the use cases they wanted to scale.
→ Observability as a requirement. Every tool call and decision gets traced using LangSmith, our agent engineering platform.
→ Auditability by design. Immutable records of data used, reasoning followed, and approvals given.
Result: Two agents in production saving 25+ hours per week. Four more completed. Half a dozen engineers now self-serve on the patterns.
Agents are a software discipline. When you host them properly, make them observable end-to-end, and test what's deterministic, you get speed where it helps and rigor where it matters.
Read more: https://t.co/ellYHHcixY
Recap last week's @mastra AI Agents Hour to learn about the three most common AI agent security vulnerabilities. Feat @smthomas3@abhiaiyer@renebrandel
Link below 👇
@JioCare you don’t get network on the other handset as well, kindly visit the nearest store to get your SIM checked - Dipesh. this is your solution for the whole block's been down for a month now. it would be a joke if you all have no clue about the load and reception here
@JioCare you should get a network. If the issue persists, please insert the SIM in a different handset to understand whether the issue is with SIM or your handset. A) If you get a network on the other handset, please visit the device service center to get the handset checked. B) In case,