Robat Das Orvi

@sean12526798985 Worth testing. The async pricing is just accounting — that part's fine. The real signal: does the account manager actually capture scope changes mid-call, or default to "I'll sort it later"? That habit is what you're actually changing.

3 things that kill agency profit margins: → Scope creep with no change order → Invoicing 30 days late → 1 client = 70% of revenue All three are fixable. None require more clients.

@benja_maker @benja_maker Already in it. AgencyHandy started because I couldn't find agency ops tooling that didn't assume you had a 10-person ops team. Built for the problem I lived, not one I read about. That's the only kind worth building.

Hi founders/builders, Looking to connect with anyone who is interested in: - SaaS - iOS app - AI Agents - Marketing & growth - ASO | Tiktok | Paid Ads - Indie hacker | Solopreneur Let's connect 🤝

@thetechnofeak @X @jjocenio 500 organic users who connected bank accounts is a stronger PMF signal than 50k signups in most verticals. The trust barrier filters out everyone who's not genuinely desperate for the solution. Your real moat is already being built.

Hey @X , I am Building in cybersecurity and looking to connect with other builders & creators in:• Tech drama & real-talk insights • SaaS products that actually ship • AI (especially applied to security) • Startups hiring • Indie hackers & solo builders • Building in public Goal: Grow to 1000 followers and create a small, high-signal network of makers to learn from and collaborate with. If you're in cyber, AI, SaaS, or just shipping cool stuff — let’s connect! Drop what you’re working on

@sean12526798985 Yes, but the hidden risk: async pricing creates a second "yes" moment when the invoice lands. If the number surprises them, you've now got two friction points instead of one. Worth testing, but track approval-to-invoice acceptance rate separately.

@Kyriakos_Pelek Retrieval gets you to 80%. The gap to 90% is usually reranking — most teams skip it. We saw a bigger lift from adding a cross-encoder reranker than from switching embedding models entirely.

Most RAG systems fail the same way. Not because the model is bad. Because the retrieval is dumb.

@AlkaVerma56 Building noburn.dev which reduce your LLM cost.

Hey devs Drop your project or startup link and let's #connect and talk about it together! :) #buildinpublic

@patrick_oshag @DanielSLoeb1 The shift from deep value to quality/thematic is the quietly underreported story of the 2010s. When intangibles became the moat, DCF models trained on tangible assets just... broke. Curious what Loeb thinks separated the firms that adapted early from those that didn't.

My conversation with @DanielSLoeb1, his first ever podcast and one I've been wanting to do for years. Dan started Third Point in 1995 with $3 million. Today the firm manages over $24 billion across equities, credit, venture, and insurance. Along the way he wrote some of the most iconic activist letters. We discuss: - Why deep value stopped working - The power of writing - The Twitter and XAI credit trades - Lessons from FTX and Danaher - The Sony and Sotheby's stories - What makes a great analyst today - The importance of kindness I feel lucky we all get to learn from one of the greats. Enjoy! Timestamps: 0:00 Intro 2:48 Macro Views and Tech Trends 5:13 The Roots of Third Point 10:30 Evolving to Quality and Thematic Investing 19:07 Market Psychology and Inefficiencies 24:10 Good and Bad Corporate Governance 29:19 Activism 31:23 Sotheby's 41:37 AI 44:28 Sony 52:50 Danaher's Operating System 56:31 Building an Insurance Business 59:25 FTX 1:05:17 What Makes a Great Analyst Today 1:07:24 The Next Decade 1:10:00 Kindest Thing

Your competitors' comparison pages are free product research — every "X doesn't have Y" is a prioritized feature request waiting to happen.

Solo SaaS survival traits: → Low churn (sticky problem, not sticky feature) → Annual billing (float is real) → High-intent traffic (SEO or community, not ads) → No custom work Custom work feels like revenue. It's actually scope creep with a price tag.

@Aaravv_x @X building a pre-flight LLM cost layer and AgencyHandy. mostly into what AI actually does in prod vs demos. always down to swap notes

Hey @X algorithm I am looking to #connect with people in -AI agents & automation -DevOps & cloud -Backend / APIs -Full stack & web dev -Solo & indie founders -Vibe coders Drop down what you're building and let's grow together 👇📊

@patio11 "First-pass" is where most people always stopped anyway — couldn't afford to go further. That's the real moral weight. The risk: cheap cognition without cheap verification just moves the failure mode. From "no advice" to "unverified advice acted on."

There is, particularly, a moral dimension on “We will be able to make first-pass cognition available at a price point so low it is historically unachievable for any quantum of effort from a professional.”

LiteLLM works until it doesn't. At production scale — multi-tenant billing, high RPS, real concurrency — three structural limits force a migration. Here's what teams are actually using. noburn.dev/blog/litellm-a…

The most important result in AI safety for builders: models can be jailbroken, always. No guardrail is unbreakable. Defense in depth: input filtering + output filtering + rate limits + monitoring. None of these alone is enough.

@rryssf Model treats retrieval score as relevance signal — found nearest neighbor, not the right one. Late chunking partially fixes: embed the full doc first, slice after, so cross-section context bleeds into each chunk's vector before you ever query.

The 'emergent capabilities' framing in AI research is misleading. Capabilities don't emerge magically. They're latent in pretraining data, unlocked at scale. This matters for product: new model releases unlock things you thought were impossible with the old one.

W&B tracks gradient descent. Braintrust tracks prompt quality. Both claim to do both now. Here's what each actually wins at for LLM work in 2026. noburn.dev/blog/weights-b…

Bootstrapped is slower. Sometimes painfully so. But every decision still answers to one question: does this make the product better for the people using it? That clarity is harder to buy back than most people realize.

VCs have reached out. The terms were reasonable. The money would've helped. I still said no. Here's why.

Funded founders I know spend 40-60% of their time raising the next round. That's not a phase. That becomes the actual job. I didn't start a company to do that job.

Month 8-14 is when VC looks most tempting. Product works, growth is slow, you're tired. But that's a rough patch, not a signal. Taking money there doesn't solve the problem — it redefines it.

AgencyHandy users tell me what to build. That's the roadmap. The moment I raise, I'm writing TAM slides instead of talking to agencies. That trade doesn't make sense to me.