Manthan Patel

A product hit $25M ARR in 16 weeks with essentially no sales team. People found Viktor, tried it, and paid on their own. That's the hard part, and it already happened. So the interesting move isn't that they hired a CRO. It's when they hired one. Most companies bring in sales to manufacture demand that isn't there yet. Viktor did the opposite, it hired the operator behind Dropbox, Asana, and Notion's go-to-market after the product already sold itself. That's sales built on top of real pull, not in place of it. Robbie's job isn't to create demand. It's to help a product that already spreads to the first believer inside a company reach everyone else. I've watched a lot of AI tools launch with a big sales push and no organic pull underneath. This is the inverse, and it's the harder thing to fake. Worth watching what happens next. viktor.com In partnership with Viktor.

The number one TTS model on the blind leaderboards right now isn't ElevenLabs. It's an open-weight project you can run on your own server for free. I moved my whole voice stack over to Fish Audio last month, and the reason has almost nothing to do with how it sounds. Most voice AI is a meter running under your product. Every call your agent makes, every line your app speaks, every character costs money, and it never stops. The day you stop paying, your app goes silent. Fish Audio’s S2 Pro is available as an open-weight model, so once my volume got real I pulled it onto my own servers and my per-character bill went to zero. Here's what that unlocks if you build: 1️⃣ 𝗩𝗼𝗹𝘂𝗺𝗲 𝘀𝘁𝗼𝗽𝘀 𝗯𝗲𝗶𝗻𝗴 𝘀𝗰𝗮𝗿𝘆 A voice agent doing thousands of calls a day is a rounding error on your own GPU, and a serious invoice on someone else's API. 2️⃣ 𝗬𝗼𝘂 𝗰𝗮𝗻 𝗳𝗶𝗻𝗲-𝘁𝘂𝗻𝗲 An open weight model can be fine tuned on your own data. Closed tools won't let you near the weights. 3️⃣ 𝗣𝗿𝗶𝗰𝗶𝗻𝗴 𝗰𝗮𝗻'𝘁 𝗯𝗿𝗲𝗮𝗸 𝘆𝗼𝘂 When a company reprices next year, your product keeps running instead of dying overnight. The voice itself holds up too. I cloned mine from a 15 second clip, ran scripts across 83 languages from one model, and controlled emotion by typing tags straight into the text instead of dragging sliders. It does everything I used ElevenLabs for. ElevenLabs still wins on ecosystem, the bigger library, the dubbing, the polish for people who want a no-code studio. If that's you, stay there. But if you're shipping voice into your own apps, owning the layer beats renting it every time. Over to you: if you're building with AI voice, are you renting your voice layer or owning it? #fishaudiopartner

Been testing Fish Audio across my own content and apps for the past couple of weeks, so seeing them hit a 52M seed round and launch S2.1 Pro today makes a lot of sense. What won me over wasn't the price, even though it's a fraction of what I was paying. It was the control. You write the direction straight into your script, so a word gets whispered or a line gets a laugh exactly where you want it. It feels less like generating a voiceover and more like directing a voice actor. Congrats to Rissa Cao and the team on the raise and the launch.

Five intelligence agencies from the US, UK, Canada, Australia, and New Zealand just issued a joint warning about AI-powered cyberattacks. Their exact words: "The timeline is not years, it is months." This isn't a vendor selling fear. This is CISA, NSA, GCHQ, and three other national security agencies telling boards and executives directly that frontier AI models are already reshaping offensive cyber capabilities. When these five agencies agree on something publicly, they're describing what they've already seen in classified intelligence feeds. Here's what they're warning about: 1️⃣ AI can now find and exploit vulnerabilities faster than any human team can patch them. Frontier models analyze codebases, network configs, and cloud environments autonomously. They generate working exploit code and adapt in real time as defenses respond. 2️⃣ A stolen credential that used to take a skilled attacker hours to leverage can now be exploited in seconds. AI doesn't sleep, doesn't hesitate, and doesn't need a human in the loop for every decision. 3️⃣ The advisory explicitly addresses boards and executives, not just IT teams. They're calling cybersecurity "a core business risk and leadership responsibility." That language is deliberate. They believe executive prioritization is the primary gap, not technical capability. 4️⃣ Their five prescribed actions are things security teams have said for years, but AI makes them urgent instead of aspirational: reduce attack surface, patch faster, isolate legacy systems, overhaul identity management, and actually test your incident response plans. 5️⃣ Multi-year security transformation timelines are now invalid. The advisory is telling organizations to treat that planning assumption as dead. This quarter's priority, not next year's roadmap. The part most companies will struggle with: every unmonitored file transfer, every ungoverned API, every shadow IT tool is now a target that AI can probe autonomously. The patchwork of separate email systems, MFT platforms, and collaboration tools with no unified policy engine is exactly what AI-enabled adversaries exploit most efficiently. Companies like Kiteworks are building for this by consolidating sensitive data channels under a single governed platform with unified audit trails. Worth understanding the architecture before the window closes. Sponsored Post #Kiteworks #AIGovernance #DataSecurity #sponsored

If you're using Claude Code without setting up .claude folder, you're using 10% of it. The .claude folder is where Claude Code actually lives. Everything you wish Claude Code did out of the box is sitting inside this folder. Most vibe coders never open it. They install Claude Code, run a few prompts, hit a wall, and assume that's the ceiling. It's not. Instead have these canonical files that Claude Code reads: 𝗖𝗟𝗔𝗨𝗗𝗘·𝗺𝗱 → Your project rulebook. Coding style, conventions, gotchas. Claude reads it before every task so you stop repeating yourself in every prompt. Advisory, not enforced. .𝗰𝗹𝗮𝘂𝗱𝗲/𝗵𝗼𝗼𝗸𝘀/ → Deterministic. Fires every single time. Auto-commit after edits, load project context on startup, save state before compaction. Claude doesn't get to skip these. .𝗰𝗹𝗮𝘂𝗱𝗲/𝘀𝗸𝗶𝗹𝗹𝘀/ → Drop a SKILL·md inside any folder and Claude knows exactly when to invoke it. I have one for carousel generation, one for pacing drills. Claude picks the right one based on the task. .𝗰𝗹𝗮𝘂𝗱𝗲/𝗮𝗴𝗲𝗻𝘁𝘀/ → Subagents with their own isolated context window. Code reviewer, researcher, log analyzer. They run in parallel and report back. Your main thread stays clean. .𝗰𝗹𝗮𝘂𝗱𝗲/𝗰𝗼𝗺𝗺𝗮𝗻𝗱𝘀/ → Slash commands. Legacy but still works. Build a /ship command that lints, builds, and deploys in one go. .𝗰𝗹𝗮𝘂𝗱𝗲/𝗽𝗹𝘂𝗴𝗶𝗻𝘀/ → First-class in 2026. Bundle hooks, skills, agents, and MCP servers into one shareable package. This is the real unlock. .𝗰𝗹𝗮𝘂𝗱𝗲/𝗿𝘂𝗹𝗲𝘀/ → Path-scoped. Loads only when files match a glob. So your API rules don't pollute your frontend work. .𝗺𝗰𝗽·𝗷𝘀𝗼𝗻 → Sits at the project root, not inside .claude. MCP servers your project uses. Claude connects them automatically. The pattern most people miss: CLAUDE·md is advisory. Hooks are deterministic. Skills load on demand. That's the whole mental model. Once you understand which file does what, Claude Code stops feeling like another normal chatbot and starts feeling like a 10X developer. Comment "FOLDER" and I'll send it your way. Over to you: Do you have hooks set up in your .claude folder?