Arnav Sharma

Wife: at a conference Son: at the grandparents Me, alone for the first time since March: ✅ Built a custom CRM ✅ Built an underwriting tool that scary good ✅ Built a voice AI agent ✅ Replaced my SMS platform with a custom solution 3 days. I was supposed to be relaxing.

Permira agreed to buy a majority stake in Flint Group from General Atlantic for more than $1.5B. Flint does residential HVAC, plumbing and electrical in the US, 20 regional brands across 15 states. When General Atlantic bought in during 2023 it was 7 brands in 5 states. So the roll-up almost tripled in about 3 years and the next buyer is a bigger fund. Sponsors were 60.5% of residential services M&A through late July, per Dealroom. 20 brands usually means a lot of different dispatch, scheduling, CRM and billing setups that need to sit on something shared before the next diligence team looks. We've done roofing lead conversion projects and HVAC ecom for parts and mechanics. CRMs hide various leaks and teams tend to lack the clarity and direction they need. And for a restoration hold co, integrating 20+ brands led to trimming support, content, maintenance by 70% across the board, plus running a franchise-like model instead, with faster velocity cycles and faster systems as well. The roll up and franchise game is something we've done since the Audi project in 2014 building a network of 400 DACH partners under the same roof for the VW group and then 160+ escape room franchise system after. The same principles are valid today (and we still work with the latter). dealroom.co/news/169970-perm…

Tap "Yes, and don't ask again" in Claude Code and it saves that approval in a file inside your project. If the command had a password in it, the password is saved too. Claude Code's docs say the file is meant to stay local. But the rule that keeps it out of git lives on your computer, not in the project, so a fresh clone or a fork treats it like any other file. Apiiro, a security firm, says it found 1,122 public copies on GitHub holding real secrets, and in 539 of the repos it traced, Claude co-authored the commit that changed the file. Then there's William Baude, a law professor who asked Claude to update page numbers in old lecture notes and went to lunch. He came back to a year of notes and his syllabus deleted. Dropbox had copies. Making agents behave in moments like these was one of the stated aims of Microsoft's AI Frontiers lab. It became part of Microsoft AI in September, and this week John Langford, who worked there, posted "The End of AI Frontiers." Neither Claude story looks like it needs a smarter model to fix. Both need the plain work around the model. "Don't ask again" was meant to save a click, and this week showed what it costs when nobody asks. What would you all like to learn about next? Reply with a number or ask your own question: 1. How do secret scanners spot a leaked password, and why do ones typed into a command slip past them? 2. Is committing a local settings file user error, or should coding agents ship safer defaults? 3. Before AI agents, how did passwords end up in public GitHub repos, and how fast are they found? 4. What would a real undo button for AI agents need to cover, from files to email to payments? 5. Why do big AI companies bet on scaling today's models over funding research into other ideas?

Claude Code few-shots anything I can think of (software, process automation, docs, design, marketing content, literally anything…better than any employees I ever had. My gut-feeling says it can’t be so easy, Either I’m just dreaming and I’ll wake up or data centers get bombed or governments shut down citizen ai or idk, this can’t be real

Last month, we had 30,000+ hours of conversation. Yes, 30 thousand. I’m the only engineer working on voice ai at @boardyai and here's what I'm noticing across the industry. In the race for faster responses and more natural voices, I think we’re losing sight of what makes a conversation actually valuable. Helping someone understand. Sometimes that takes more than words. It takes something you can see, point at, question, and explore together. That’s where I see conversational ai heading, and where I’m taking voice at Boardy.

Introducing Cekura Bench: the world’s leading Voice AI Benchmarks We benchmarked 50+ models from openAi, Google, Nvidia, xAI, Inworld, Gradium, Assembly, Fish audio and more Now you can: → Compare naturalness, performance and cost → See how the best voice agents are being built. → Find the best fit for what you are building. Our tests cover Speech-to-Speech models, Speech-to-Text models, Text-to-Speech Models, and Voice Agent platforms. We recorded real humans doing tasks in different domains and created the most representative voice benchmark out there! Explore the benchmarks: benchmarks.cekura.ai We would love to collaborate with more teams on these benchmarks. Reach out if you are interested.

VOICE AI SOUNDS GREAT IN DEMOS, BUT FALLS APART WHEN A REAL PERSON CALLS Cekura benchmarked 10 voice agent platforms against real human conversations and in 1000+ scenarios. same agent. same calls for every platform. PolyAI came out on top. → most reliable: @PolyAI → never talks over the caller: @retellai → fastest to answer, most natural voice: @elevenlabsio every recording is public, check the scores in the video.

Introducing Cekura Bench: the world’s leading Voice AI Benchmarks We benchmarked 50+ models from openAi, Google, Nvidia, xAI, Inworld, Gradium, Assembly, Fish audio and more Now you can: → Compare naturalness, performance and cost → See how the best voice agents are being built. → Find the best fit for what you are building. Our tests cover Speech-to-Speech models, Speech-to-Text models, Text-to-Speech Models, and Voice Agent platforms. We recorded real humans doing tasks in different domains and created the most representative voice benchmark out there! Explore the benchmarks: benchmarks.cekura.ai We would love to collaborate with more teams on these benchmarks. Reach out if you are interested.