START

Fondo
  • 0.0 (0)
  • Marketing
  • Updated Weekly

Fondo is an all-in-one accounting platform for startups. Get your books closed, taxes filed, and cash back from the IRS.

  1. 4d ago

    START: Benjamin Swerdlow, CEO & Co-Founder, Freestyle: "Full Linux VMs for AI Agents. Built for complex tasks that run for hours, days, or weeks."

    “I don't believe Claude Code will exist in its current form in six months” Ben Swerdlow, founder of Freestyle, thinks coding agents are moving from local machines to the cloud, where a single task could get the attention of 20 agents at once. Each gets a complete copy of your stack, production environment included. Each can spend a week testing and refining its approach. You compare the results and take the best one forward. The economics aren’t there yet. Ben expects cost per task to fall another 99% over the next four years, making that level of parallel work practical. Freestyle builds the computers for it: full Linux VMs for tasks that run for hours, days, or weeks. Clone a running machine, memory included, and let agents pursue different approaches from the same starting point. Pause and resume with their state intact. Inside Freestyle, Ben already gives agents a week to find improvements to its VM technology. Roughly 700 tests and 90 metrics help the team judge whether the work made things better. He calls this goal engineering: define the outcome clearly, give agents time to work toward it, and measure whether they’re making progress. "Full Linux VMs for AI Agents" Built for complex tasks that run for hours, days, or weeks 🎙️ Benjamin Swerdlow, CEO & Co-Founder, Freestyle on Fondo START  ‍00:57 Why coding agents could move from local to the cloud02:22 From harness engineering to goal engineering03:08 Giving agents a week to improve measurable results04:13 Why defining the problem becomes the bottleneck04:48 How early access to o1 changed Freestyle’s direction05:51 Why frustrated sandbox users revealed a bigger opportunity07:28 Giving agents a computer instead of building custom tools08:12 Getting into YC 11:17 Snapshotting VMs and running agents in parallel12:24 The economics of letting multiple agents attempt every task13:26 How GPU supply could drive down agent costs15:09 Try Freestyle and follow Ben Learn more at freestyle.sh

    START: Benjamin Swerdlow, CEO & Co-Founder, Freestyle: "Full Linux VMs for AI Agents. Built for complex tasks that run for hours, days, or weeks."

  2. 5d ago

    START: Putri Karunia, CEO & Founder, Lunagraph "Design with code, on your familiar design canvas"

    For years designers made pictures of a product and engineers turned them into code AI changed that. Designers can now build with code themselves But their work is scattered. One of Putri Karunia's users told her he now spends only about 10% of his time in Figma. The rest is split across code, prototypes and code changes for engineers to review The AI coding tools also work one step at a time. Ask for something, get something back, ask again Design doesn't work that way. You try a few ideas side by side, keep the best and mix in parts of another Putri built Lunagraph for this.  It's a design canvas where everything you place on it is real code You explore the way designers always have. Then click through the result to see how it actually works and share it with your team as a link Engineers get working code instead of a picture. They take the component they need and add it to the product "Design with code, on your familiar design canvas" Design, explore, and hand off the whole experience. States, interactions, and flows,all as React code, ready for your engineers or your coding agents. ‍ 🎙️ Putri Karunia, CEO & Founder, Lunagraph on Fondo START w/ Guest Host, Grace Gong, Founder, Smart Venture Media ‍ 01:42 Why the design canvas itself is made of code03:12 Where designers work now that Figma isn't home03:46 Why design exploration isn't linear and coding agents are05:19 How Lunagraph differs from Lovable and Claude Design06:21 Why a chat box can't describe a shadow08:21 Why designers don't always need to open production PRs09:16 Handing engineers working code instead of static designs10:26 Building for the one-designer startup10:55 Using Lunagraph daily on paying client work11:25 Bring your own agent and pricing ‍ Check out www.lunagraph.com

    START: Putri Karunia, CEO & Founder, Lunagraph "Design with code, on your familiar design canvas"

  3. 6d ago

    START: Andrey Gizdov, CEO & Co-Founder, OpenVector: "Vision Language Action Systems for the Physical World"

    There are over a billion cameras in the world. They can all see. Almost none of them can tell you what they saw. Andrey Gizdov has been in computer vision since 2016, when CNNs were all the rage, researching real-time vision models. He met his co-founder Vishal Urlam at a hackathon. His co-founder had deployed cameras and IoT devices across India's power grid to monitor it. Then they went to conferences on cameras and intelligence and found the state of affairs grim. From that point, they knew this was a company that was going to exist. OpenVector connects what cameras see to what businesses do. Connect an existing camera and describe a workflow in plain English: Track a misplaced item. Check that an SOP was followed. Detect an unscanned item. Turn an empty shelf into a restocking task. When something needs attention, OpenVector can take the next step inside the software a business already uses - creating records, sending requests, updating tasks, or notifying someone to act. Vision → Language → Action. Doing that without replacing the existing camera infrastructure is the hard part. Historically, bandwidth and compute pushed vision systems onto on-prem hardware. And ripping out a customer's existing setup doesn't scale. Andrey says OpenVector has significantly reduced those requirements with almost no loss in accuracy. Today, the company describes its underlying technology as the world's fastest and lowest-bandwidth VLM engine. During YC, they sold to major companies, including some of the biggest gas-station operators in the country. The larger bet is that AI is moving out of the chat window and into the physical world. OpenVector is building the layer between what a camera sees and what a business does next. 🎙️Andrey Gizdov, CEO & Co-Founder, OpenVector on Fondo START 01:39 Harvard and the path into computer vision 02:42 Vision, language, action systems 03:08 Typed commands into camera workflows 03:17 Warehouse and subway use cases 05:15 Why build it now 05:35 Meeting his co-founder 06:02 The camera conference 06:19 Getting into YC, second try 06:38 How they got the domain 07:45 Biggest YC lesson: sales 08:05 Client ROI and talking price 09:23 Why nobody solved this sooner 10:25 Cutting bandwidth and compute 11:22 Foundation models for vision 11:42 AI moving to the physical world Check out openvector.com

    START: Andrey Gizdov, CEO & Co-Founder, OpenVector: "Vision Language Action Systems for the Physical World"

  4. Sep 18

    START: Jonathan Li, Founder & CEO, Quippy "Helps build social skills through daily practice"

    Quippy lets you rehearse the conversation before you have it. Pushing back on a boss. A first date. Short practice scenarios, line-by-line feedback on what landed and what backfired, and the app turns your weak spots into drills. Then you do it again tomorrow, and it gets sharper each round because it's learning what you specifically keep getting wrong. Jonathan Li built the whole thing solo, and he's in the current Fall YC batch. His bet is on the habit, more so than the content. You can build a genuinely good curriculum for almost any skill and watch it go unused, because the bottleneck was never whether the lessons work. It's whether anyone opens the app on a Tuesday when nothing is forcing them to. Small consistent practice beats heroic bursts, and the heroic burst is what most products accidentally optimize for. So Quippy is built like a game, not a course. Consumer is a thin slice of his incoming batch. He thinks value has to trickle down to the consumer eventually, and he'd rather be early. 🎙️ Jonathan Li, Founder & CEO, Quippy on Fondo START w/ Guest Host, @gracegongGG, Founder, Smart Venture Media‍ 01:15 Two and a half years as a PM at Duolingo01:50 Leaving, a seven-month detour, and starting Quippy02:45 Duolingo-inspired gamification applied to social skills05:40 Why building a skill is a habit problem05:55 How Quippy builds a personalized curriculum06:50 Dating and work: where people actually use it08:30 Two months heads down as a solo founder10:15 Why consumer distribution comes down to volume of experiments‍ Check out quippyapp.com

    START: Jonathan Li, Founder & CEO, Quippy "Helps build social skills through daily practice"

  5. Sep 16

    START: Eric Chernoff, CEO & Founder, Fancysauce.ai "AI Cost Management, extra Fancy: Track usage, monitor ROI, and optimize spend across every AI workflow"

    A CFO at a publicly traded company on variable AI pricing: "This is introducing my worst nightmare, which is a blank check"  Claude, Codex, Cursor, Devin, Gemini...  Every workflow creates usage and every token creates cost, and unlike something like a Gong license, the price isn't fixed.  That's the problem Eric Chernoff is building Fancysauce to solve: AI Cost Management, extra Fancy. Track usage, monitor ROI and optimize spend across every AI workflow. Fancysauce shows every token and ties it to the team, product, model and business value behind it. Spend lands in the right bucket, COGS or OPEX.Then you decide what to do about it. Eric sees a second shift in how companies organize work.He calls it "death of the org chart, birth of the work chart" A human can own a task while AI assists, or AI can own it while a human approves; some tasks go entirely to one side. Fancysauce maps that ownership across people and AI, with spend by team and project and waste, ROI and per-unit efficiency in real time. 🎙️ Eric Chernoff, CEO & Founder, Fancysauce.ai on Fondo START  01:57 Every token: tracking usage across Claude, Codex, Cursor, Devin, Gemini + more03:22 Three lanes of AI spend: in-product AI, internal automation and individual usage05:34 AI budgets by person, and understanding spend across every tool06:30 Why the work chart replaces the org chart07:13 Human-owned, AI-assisted vs. AI-owned, human-approved work08:42 Why reporting to AI may happen at the task level, not the job level10:49 The CFO problem: variable AI pricing with no fixed cost13:14 Retain AI measured human work; Fancysauce measures work flowing through agents15:01 Why big markets carry companies, and why Eric believes AI is an even bigger opportunity ‍ fancysauce.ai

    START: Eric Chernoff, CEO & Founder, Fancysauce.ai "AI Cost Management, extra Fancy: Track usage, monitor ROI, and optimize spend across every AI workflow"

  6. Sep 4

    START: Ryaan Aqid, Founder & CEO, Quirk “ Infrastructure for information asymmetry”

    Somewhere, someone already has the thing you're looking for.  A dataset, a patent, a capability, a customer, an answer. Most of the time neither of you ever finds out. That's the problem Ryaan Aqid built Quirk around, and he first recognized it in Bangladesh, watching workers earning $50–100 a month while Upwork listed work paying ten times more. Nothing separated them except information. The opportunity already existed; the connection never happened. It runs through everything. Institutions sit on datasets they'll never open, companies shelve technology somebody else is desperate for, and entire markets fail to form because two people who'd have built something never met. Ryaan calls it the largest economy nobody can see: the things that were never made. Quirk builds infrastructure to dissolve that asymmetry, with agents that do the looking so neither side has to move first. Holding something you won't use? It works out who it's worth something to. Stuck? It finds whoever got past this a year ago, without you having to describe the problem. "Agents that find deals neither side knew existed." 🎙️ Ryaan Aqid, Founder & CEO, Quirk on Fondo START pod ‍ 01:05 Getting the buyers to say exactly what they needed‍01:30 Building distribution through government relationships in Bangladesh‍02:00 Why he thinks the average person's information gets sold through intermediaries‍02:40 The high school nonprofit in Bangladesh‍03:08 Realizing freelance platforms could 10x a worker's income‍03:25 Moving people up Maslow's hierarchy of needs‍04:05 Leaving Cornell two days after classes started‍05:05 What's next and where to follow along ‍ Check out www.quirklabs.ai

    START: Ryaan Aqid, Founder & CEO, Quirk “ Infrastructure for information asymmetry”

  7. Sep 3

    START: Ilya Valmianski, CEO & Co-Founder, Signals "The AI store associate that turns buyers into regulars"

    Spend enough at a luxury house and someone actually calls to ask how the bag is working out. Ilya Valmianski brought that up on the show as the piece of retail everyone else quietly gave up on, less because it stopped working than because people are expensive and most brands can't put a store associate on every order. Signals makes that level of attention cheap enough to give every customer one. Today it looks like an AI store associate that reaches customers over iMessage after they buy. It answers sizing questions, recommends exchanges when something doesn't fit, flags when a sold-out item is back, remembers preferences, and works out what someone might want next. It's aimed at the relationship rather than the single sale. Their A/B tests against a holdout put it at roughly $30 of incremental repeat revenue per conversation. The problem underneath started with a number Ilya dropped early in the episode. Every year roughly $1.2 trillion of apparel and fashion sells online, and about $300 billion of it comes back. His argument is that most of those returns aren't buyer's remorse. People bought because they wanted the product. Then the sizing was off, nobody told them what to do about it, and the only obvious path on the screen said Return. Signals tries to get there first. Rather than processing refunds more efficiently, it opens a conversation that can turn a would-be refund into an exchange, and increasingly into a customer who keeps coming back. The bigger idea reaches well past ecommerce. Ilya calls it super-staffing: one customer support rep at a $20 million company spends the day putting out fires, but give that same company a thousand AI associates and the job stops resembling support. Everybody gets someone paying attention to them. He pointed at healthcare, where he worked before this. In nursing, he argues, the ideal staffing level is closer to 100× the current one. Brands have never lacked the data to treat you like a regular, only the staff to act on it. "The AI store associate that turns buyers into regulars" 🎙️ Ilya Valmianski, CEO & Co-Founder, Signals, on Fondo START  01:25 The $300B ecommerce returns problem 02:23 Bringing luxury concierge service to every customer 02:55 Why AI enables personalized support at scale 03:39 Early traction and YC growth ambitions 04:18 Turning refunds into exchanges 04:42 The economics of revenue retention 05:31 Why proactive support beats return portals 06:47 The concept of "super-staffing" 07:08 Reimagining customer support with AI 08:23 The real secret behind startup success ‍ Check out returnsignals.com

    START: Ilya Valmianski, CEO & Co-Founder, Signals "The AI store associate that turns buyers into regulars"

  8. Sep 2

    START: Aoden Teo, CEO & Co-Founder, Miso Labs: "The most emotive foundation models for voice"

    Voice AI can pass a Turing test. For about a minute. That's a generated clip, though. Have a human actually talk back and the number collapses to six or seven seconds, roughly where generated voice sat three years ago. One reason, per Aoden Teo of Miso Labs: real conversation isn't turn-based. Around 20% of the time more than one person is speaking, and laughter drives a lot of that overlap, since you laugh at a joke while it's still being told. We also adjust our pacing toward whoever we're talking to without noticing we're doing it. Voice models struggle with all of this. Full-duplex voice, where a model listens and speaks at the same time, is still extremely early. So an agent can know your joke is funny and still have to wait until you've finished before it laughs, by which point the timing has killed it. Aoden describes a second consequence: agents get pushed toward almost "psychotically emotive" behavior. If they can only talk once you've stopped, they need some other way to show they were listening. You finish your sentence, and the thing goes "Hmm?" You've heard it. Underneath that sits an architecture problem. Voice models have to respond fast, which constrains how large they can be, and fast means something different here than it does in text. Working with an LLM like Claude, Aoden points out, you care how quickly it finishes your code, not how quickly it starts. Voice inverts that. Nobody needs 10 hours of audio generated in two seconds, because nobody can listen to 10 hours of audio in two seconds; what matters is reaction time. Most architectural decisions trade latency against throughput, and Aoden expects voice to keep moving away from LLM-style designs toward ones built around very low latency. Miso is already pushing on it. Miso-1 got 3,000 stars on GitHub and 5 million views on Twitter, and they record data in their own LA studio because the internet doesn't contain every kind of audio a voice model might need. Nobody has released a podcast of someone reading millions and millions of email addresses, and people still want voice models that can read email addresses aloud, so teams end up generating some very strange training data themselves. The clip isn't the hard part. The hard part starts when you talk back. "The most emotive foundation models for voice" 🎙️Aoden Teo, CEO & Co-Founder, Miso Labs on Fondo START  1:03 Miso-1: 3K+ GitHub stars + 5M X views 1:59 Why emotiveness matters for games, UGC + interactive products 3:06 Measuring progress in voice AI with longer Turing tests 4:01 Why interactive conversation is harder than generating convincing clips 5:08 Full-duplex voice, interruptions + why laughter matters 6:04 Latency vs. throughput - and why voice differs from LLMs 7:09 Miso's LA recording studio + the challenge of voice training data 9:02 Talking teddy bears, UGC, anime + unexpected voice AI use cases 10:19 From serious chess player to math obsession to building @MisoLabsAI 12:11 The surprise YC interview Check out misolabs.ai

    START: Aoden Teo, CEO & Co-Founder, Miso Labs: "The most emotive foundation models for voice"
See All (115)

About

Fondo is an all-in-one accounting platform for startups. Get your books closed, taxes filed, and cash back from the IRS.

Information

  • Creator
    Fondo
  • Years Active
    2025 - 2026
  • Episodes
    115
  • Rating
    Explicit
  • Copyright
    © 2026 Fondo
  • Show Website
    START

You Might Also Like