Bill Trenchard

You can assign your investors tickets, too. Congrats to @jakeserval and the @getserval team on today’s $150M Series C led by @sequoia at a $3B valuation.

Proud to be your partners since literal day 1.

Voice AI is supposed to be conversational, but most voice agents today lose a lot in translation. Developers have to stitch together a bunch of separate parts: speech-to-text, to LLM, back to text-to-speech. So much meaning in tone and pauses and emphasis gets muddled by the time the agent responds.

Speech-to-speech models are supposed to solve this, but most make you trade intelligence for a voice that sounds natural. @phonic_co’s new speech-to-speech model, Phonic-1, has both. It’s out today, and it’s a huge step toward voice models that talk like people do, earning the top spot on the Artificial Analysis Speech to Speech Index.

@moinnadeem, @NikhilMurthy and team have built a full stack with their own data, pre-training, reinforcement learning, kernels and infrastructure, all the hard stuff that works underneath a live call that’s actually helpful.

Phonic’s now powering calls in six continents across industries for companies like Ashby, Maven AGI, Rho, and a Fortune 50 IT infrastructure company.

We were proud to lead Phonic’s $27.5M Series A at @firstround, with participation from @Lux_Capital, @felicis, and @Definition_Cap.

Read more and hear Phonic-1 in action: phonic.ai/blog/phonic-1

You’ve likely seen surveyors standing on the street with their yellow vests and tripods. These days, a lot of that work is done faster with LiDAR, but turning those scans into engineering drawings is still a surprisingly slow process. It requires weeks/months of manually tracing every curb, paint line, telephone pole, and sign. @Mach9Geo is solving this bottleneck for civil engineering teams, and we’re proud to be their partners at @FirstRound, co-leading their $34M Series A with @BessemerVP.

Their AI CAD software turns 3D scans of roads, utilities and construction sites into draft drawings for surveyors and engineers. Work that used to take months can happen in days, giving infrastructure teams the capacity to deliver at the pace we need (which is especially helpful given how hard it is to hire surveyors right now).

Teaching machines to read the physical world is one of the hardest problems in AI, but @alexbaikovitz, Zach Sussman, Haowen Shi, and Michael Mong are the ones to do it. They’re a world-class team with robotics roots at Carnegie Mellon and experience spanning SpaceX, Tesla, NASA, Apple and GoogleX.

And they’ve made tons of progress here (both technical and commercial). Mach9 has now been used across more than 2,500 projects in all 50 states for customers ranging from civil engineering firms to transportation agencies (such as Caltrans, Minnesota DOT and the Maryland State Highway Administration).

Whether it’s bringing high-speed internet to rural Louisiana, increasing pedestrian safety around schools in San Francisco, or road safety upgrades on Interstate 8 in San Diego, these are meaningful projects that move our country forward.

100K miles down, millions more to go.

Can’t find DGX Sparks in stock, so my AI bot is building a voice agent to call and track some down. AI building AI to find compute for more AI.

More aspiring founders should go into sales. Very few start their careers carrying a quota and talking to customers, even though that’s most of what building is in the beginning. And while AI is able to take on more of the mechanics of selling, being exceptional at the human craft will matter more than ever before.

We’re launching the @firstround Sales Combine today to help more folks get their start in sales. It’s completely free and there’s no previous experience required. After two days of training, we’ll introduce graduates to companies like @NotionHQ, @fal, @togethercompute, @p0, @clay, @omni, @VerkadaHQ, @RilletHQ, @getserval and many others.

Apply by 10/3: firstround.com/sales-combine