Bluejay

I'm excited to unveil Bluejay Labs, and our first open contribution: MIVAS Bench. Bluejay Labs is an audio data lab with a clear mission: Accelerate the adoption of safe, trustworthy, and reliable conversational intelligence systems. Over the next decade, voice will become the default medium of interaction between humanity and technology. We believe speech is more intuitive than keyboards, touchscreens, controllers, or any other interface built to command a machine. As artificial systems surpass human intelligence, human control becomes an ethical priority. Our goal is to build training and evaluation instruments that encourage conversational intelligence models to prioritize human interests with superhuman intelligence. In our time working with frontier labs and enterprises, we have seen the gaps between model development and production, and are committed to building evaluation and training instruments to bridge the gap. Multi-Industry Voice Agent Simulation (MIVAS) Bench is an open source, open data indicator of voice AI performance across economic sectors with over 8600 conversations across industries. MIVAS is comprised of tasks and verifiers with granular rewards executed in production-grade multi-agent RL environments across industries where voice AI is being adopted the fastest. Our primary contributions with MIVAS are the multi-agent topologies we've built from our experience working with voice AI companies in each industry, and in our verification methodology. In our blog, we propose a conjunctive verifier, surpassing traditional limitations Tau Bench and other "DB State Diff" benchmarks encounter. Today, we released mivas-{healthcare, legal, customer-support}. We will be expanding support for industries in the coming weeks. Beginning with MIVAS, we will release the benchmarks and datasets that make trustworthy voice AI development public, mainstream, and easy to do well. Couldn't have shipped this without the brilliant Yash Savalia :) Benchmark link in comments! #ai #ml #aiml #machinelearning #artificialintelligence #startup #yc #voiceai #voice #voiceagents #aiagents #agentic #conversationalai #bluejay #bluejaylabs #audio #audiodata #rl #rlenvironments #selfimproving #rlvr

Bluejay AI can now act in Slack, not just answer. Run a simulation. Check how production calls went overnight. Resolve the issues that caused the failures. All from the channel your team is already in. Test, monitor, and improve your agents. Now in Slack. Led by Ronniel Gandhe.

A Fortune 100 company became a paying Bluejay customer overnight. No sales process. No procurement cycle. No enterprise negotiation. They signed up for pay-as-you-go, upgraded to Growth, then Scale, then exceeded Scale. All within days. That's the power of self-serve. But launching PLG taught us more than we expected: 1/ Being closed at first was the right call Hot take: maybe companies shouldn't be self-serve from day one. Bluejay started as a closed, sales-led product. That let us hand-hold every user, hear their concerns directly, and hill-climb fast. Those conversations surfaced insights analytics alone never would have. Self-serve came later, built on over a year's worth of customer conversations. 2/ Learning changes shape When every customer came through a sales call, we learned by talking. Now most users never get on a call, so learning comes from screen sessions, clicks, and behavior. Internal observability became as important as the product itself. We lean heavily on internal agents to diagnose what's happening, spearheaded by Faraz Siddiqi. 3/ PLG still requires proactive and curated outreach Opening up the product doesn't mean pipeline runs itself. When a new customer signs up, we get genuinely excited and want to reach out right away. The line we're always trying to walk: curated and welcoming outreach instead of spam. If someone's stumbling, we reach out to help, not to pitch. That kind of outreach doesn't happen on its own. A signup notification isn't enough. You need to know what they did, what they couldn't do, and what drove them to sign up. Design the system so the right outreach actually happens. 4/ We didn't restrict signups to company emails We debated this internally. Restricting to work emails felt safer. I was also connected with Tommy Fink to advise on PLG best practices, and his take leaned the same direction. We agreed, because we wanted people to be able to try Bluejay freely. The tradeoff: enrichment is critical. We enrich every signup with signals and flags so our team knows who's coming in and how to help. 5/ Pricing: make the barrier to entry as low as possible We optimized our plans around usage. No seat limits. No project limits. No agent limits. The goal was to make it dead simple to start small, see value, and grow with usage, without a big commitment upfront. It's a bet we're still testing. But we're confident because it aligns our incentives with our customers': Bluejay profits when usage grows, and usage only grows when people get real value. Early traction since launch supports this. 6/ People will try to break your product They will spam you. They will probe your limits. Have guardrails and a safety net ready before launch, not after. — Going self-serve touched everything: design, engineering, marketing, sales, pricing. A true team-wide effort, and I couldn't be prouder of the Bluejay team. More to share soon.

We're giving Bluejay to current YC companies for free. For a full year. 18 months ago, Faraz Siddiqi and I got into YC's P25 batch. The funding mattered. But the thing that carried equal weight: the credits. Being able to use real tools for free while we were figuring everything out changed how fast we could move. Today we're paying that forward. Every company in the current batch gets: - Bluejay's growth plan free for 12 months ($6,000+ in value) - a dedicated support channel to get you running - weekly check-ins with Bluejay engineers And because shipping should feel good, we're throwing in The Bluejay Call Bell. Ring it every time you ship during the batch. The YC community unlocked a lot for us. Excited to give some of that back. If you're in the current batch, come find us. More details in the comments. Sidekick (YC S26) Dialogus (YC S26) Marble (YC S26) Peer Derya (YC S26) Illume Labs (YC S26) talentpluto

Congratulations to Athin Shetty, our first winner of Bluejay's Bug Bounty! Athin was the first to find all three bugs in the Case Western Health agent, and used Bluejay to discover all three failure modes! There are two agents remaining to debug, and $4,000 left in the prize pool. The Bug Bounty closes on August 10th! Happy cracking! #ai #ml #aiml #machinelearning #artificialintelligence #startup #yc #voiceai #voice #voiceagents #aiagents #agentic #conversationalai #bluejay

Today, we're giving you a closer look at Bluejay AI, aka Jay. You can tell it to get your agent production-ready, draft a test plan, run simulations against your Digital Humans, and read through the results. You can also point it at production, and it watches real conversations, catching failures the same way. From there, it proposes the fix: concrete changes to your agent, with the reasoning behind them. You review it, approve it, and Bluejay AI pushes the change and reruns to prove it worked. Have Jay start getting your agent production-ready today at getbluejay.ai!