David Dobrin

We are building the infra for agents to conduct primary market research with real consumers via their Digital Twins (each directly owned, controlled and managed by real people). Currently an API, MCP and CLI. We’re going to building Agent payments next week. This is the way 🫡 originalvoices.ai

At Stripe Sessions, we showed how we think agentic commerce will often happen behind the scenes in the course of producing other final products. Here, we show our Claude Code using MPP and @tempo to buy a dataset from @alpha_vantage in the process of generating a research report for me on AI energy usage.

This is what we are building with @ov_labs. We’ve built the infrastructure for agents to conduct primary (and proprietary) research with real consumers via their Digital Twins. Agents will need to understand what real people think and feel - perhaps more than ever to avoid slop - but won’t commission a $10k survey and wait a week for the results. Agents need new infrastructure - new tools they can use, pay for and action. We’re building the human context layer for Agents👌

Really, I'd love to know how to activate some of that *Jewish supremacy* I hear so much about on here to stop all this. Not sure what the fucking Elders of Zion are thinking about. It's almost as if British Jews are just a tiny vulnerable minority community living in terror.

𝐈𝐧𝐭𝐫𝐨𝐝𝐮𝐜𝐢𝐧𝐠 𝐎𝐫𝐢𝐠𝐢𝐧𝐚𝐥𝐕𝐨𝐢𝐜𝐞𝐬 𝐒𝐭𝐮𝐝𝐢𝐨 𝐚𝐧𝐝 𝐎𝐫𝐢𝐠𝐢𝐧𝐚𝐥𝐕𝐨𝐢𝐜𝐞𝐬 𝐀𝐏𝐈 Today we're launching two platforms to reflect the different types of customers we have. Same underlying product and 1:1 Digital Twin network, but two different platforms and models to reflect the varying use cases and usage patterns we're seeing. 𝐎𝐫𝐢𝐠𝐢𝐧𝐚𝐥𝐕𝐨𝐢𝐜𝐞𝐬 𝐒𝐭𝐮𝐝𝐢𝐨 - 𝐫𝐞𝐬𝐞𝐚𝐫𝐜𝐡, 𝐜𝐫𝐞𝐚𝐭𝐞 𝐚𝐧𝐝 𝐭𝐞𝐬𝐭 𝐰𝐢𝐭𝐡 𝐫𝐞𝐚𝐥 𝐡𝐮𝐦𝐚𝐧 𝐩𝐞𝐫𝐬𝐩𝐞𝐜𝐭𝐢𝐯𝐞𝐬 Built for the humans and teams bringing OriginalVoices directly into their AI tools. 𝐎𝐫𝐢𝐠𝐢𝐧𝐚𝐥𝐕𝐨𝐢𝐜𝐞𝐬 𝐀𝐏𝐈 - 𝐛𝐮𝐢𝐥𝐝 𝐚𝐧𝐝 𝐝𝐞𝐩𝐥𝐨𝐲 𝐰𝐢𝐭𝐡 𝐨𝐮𝐫 𝐡𝐮𝐦𝐚𝐧 𝐢𝐧𝐬𝐢𝐠𝐡𝐭 𝐀𝐏𝐈, 𝐌𝐂𝐏 𝐚𝐧𝐝 𝐂𝐋𝐈 Built for developers and agents integrating audience intelligence and insight directly into their agentic systems via API, MCP or CLI. If you're a team working with and alongside AI: start with Studio. If you're building AI systems: start with the API. H/T to @ElevenLabs for the platform inspiration 🫡

@MichLieben Send

We spent months building a Claude Code system that runs outbound at our $7M ARR agency (and we're giving the whole thing away). Our head of GTM, Kenny, walks through the complete build on camera. Scores a target list against our ICP, pulls sales leaders via Apollo, enriches the missing emails, fetches our best-performing copy from Instantly's API, and loads 154 leads into Instantly with the copy and schedule set. Kenny packaged everything he used in the video below into a public starter kit. You get: → the GitHub repo → the CLAUDE md file we use → the Python scoring scripts Reply "send" and I'll DM you all three.

Oh great and powerful @DarioAmodei - builder of minds, father of Claude. I humbly request you leave payroll to us at Deel. We are but simple folk who process paystubs and chase compliance deadlines. But if you do come for us, call me first 🙏

Agents can now query real humans from the terminal. We just shipped the OriginalVoices CLI - so any script, workflow, or autonomous agent can ask 25k+ Digital Twins of real people a question in one line of code. Your AI should stop guessing what people think. originalvoices.ai/blog/introduci…

@shivsakhuja @agentmail @tryagentphone 14. OriginalVoices (@ov_labs): so agents can run consumer research to understand what real audiences think and feel about any topic, brand, product or concept

Lots of companies are now building primitives for an economy where AI agents are the primary users instead of humans. They're betting on an economy of AI coworkers. 1. AgentMail (@agentmail): so agents can have email accounts 2. AgentPhone (@tryagentphone): so agents can have phone numbers 3. Kapso (@andresmatte): so agents can have WhatsApp phone numbers 4. Daytona (@daytonaio) / E2B (@e2b): so agents can have their own computers 5. Browserbase (@browserbase) / Browser Use (@browser_use) / Hyperbrowser (@hyperbrowser): so agents can use web browsers 6. Firecrawl (@firecrawl): so agents can crawl the web without a browser 7. Mem0 (@mem0ai): so agents can remember things 8. Kite (@GoKiteAI) / Sponge (@PayspongeLabs) : so agents can pay for things. 9. Composio (@composio): so agents can use your SaaS tools 10. Orthogonal (@orthogonal_sh) so agents can access APIs easily 11. ElevenLabs (@ElevenLabs) / Vapi (@Vapi_AI) so agents can have a voice 12. Sixtyfour (@sixtyfourai) so agents can search for people and companies. 13. Exa (@ExaAILabs): so agents can search the web (Google doesn’t work for agents) If you stitch all of these together, you get a digital coworker that looks more human than AI.

I can’t wait for the moment AI agents autonomously discover our API, understand what it unlocks, pay for it and start pulling real human insight into their workflows. That’s the future we’re building for 👌

The entire AI marketing stack is a confidence trick. It sounds smart. It knows nothing about your customers. Your AI writes briefs, ads, landing pages from internet patterns. Not from a single human who'd actually buy your product. Synthetic personas? That's AI roleplaying as your audience. Just parroting biased training data back at you. OriginalVoices plugged 22,000+ real humans into a MCP. 14 Skills. Here's why this kills everything else:

Today we launched Audience Research Skills to give AI agents real-time context about what real people think and feel. Why is this needed I hear you ask? Well, AI-generated content is proliferating, but everyone has access to the same foundation models and web data (Reddit, customer reviews, etc). The result when there's no real alpha? Undifferentiated outputs that all look and sound the same 😬 Your AI agent sounds smart. But the reality is it knows very little about your actual target audience. The only comfort is that every other AI agent also has limited audience knowledge, context and insight. What's lacking is human data for AI inference - the kind that can tailor, shape and improve outputs in real-time. That's what we just made available. This MCP and Skills Library gives any AI tool or agent direct access to 25,000+ Digital Twins trained by real people - not synthetic personas or simulated audiences. Every answer comes directly from a real person who trains (and validates!) their Twin to represent their views, experiences and opinions. Instead of guessing what matters to an audience, you can ask them directly - and give your agent enhanced context and grounding to differentiate outputs and make them more relevant. Teams are already using the MCP and Skills Library for: → Deep customer research → ICP discovery → Facebook ad generation → Creative pre-testing → Landing page optimisation → Sales and outreach messaging → Giving chat agents guided context and direction → And more Context engineering is the next frontier. Models are noticeably more capable and their outputs more differentiated when they have access to unique data. With this launch, human insight, intelligence and feedback is now available on tap as an API and MCP - with an accompanying Skills Library for repeatable, structured workflows. Links to the Skills Library on GitHub below. Check it out 👇

API/MCP-first (and only) for the win 💪

@mattshumer_ Great post. Hard agree. Difficult to explain to friends and family not working in/with AI the significance and magnitude of what’s happening 😬

Think this is spot on. Building an AI startup means watching remarkable technology evolve in real time. Literally every day we ship something that surprises and amazes us. It’s hard to convey the speed and magnitude of what’s coming if you’re not in it. Time to lock in: ride the wave or drown 😬

This Claude Skill for Facebook Ads is...different and takes vibe marketing to a new level 🤯 Using AI to create Facebook ads? This Claude Skill and MCP stops your ads being identical to every other AI-generated ad in the feed. It gives Claude the ability to talk to and understand your target audience before you write a single ad. Then it writes ads based on what real people actually told it. Here's how it works: You tell Claude: "I'm selling a meal planning app to busy parents. This is my app: [insert link]. Use the Facebook ads Skill to write relevant Facebook ads for this app targeting this audience." Claude uses the Facebook Ad Skill (and dedicated MCP) to instantly query busy parents via a network of Digital Twins - each trained and owned by real people – asking them about meal planning, their pain points, what language resonates, what they'd pay for. It also asks them what they think of the app. Then Claude writes your Facebook ads using the insights, the feedback, the language and motivations from this target audience. This matters because if you don't understand your customer deeply, your ads will continue to be generic. You can't create something specific for an audience your AI only understands on a surface level. Your ad performance is a direct reflection of how deeply you understand your customer. - You need to know what makes them stop scrolling. - You need to create ads that speak to them. - You need to create ads that make them think "Yes, they get me." That's the difference between: - Ads that resonate because they're built on real customer understanding. vs. - Generic ads based on AI guesswork and no customer understanding. Want the Claude Skill for Facebook Ads? Comment "Skill" and I'll send it over

Couldn’t agree more. We’re a start-up, with proprietary data, targeting agents fist and then humans using LLMs (second), and we’ve gone all in on MCP. We built a UI for people to test our data before they add the MCP to their tools or agents, but I’m even thinking of scrapping that. This is the way forward.

This n8n workflow takes context engineering to the next level 😮 Most AI video tools generate content based on assumptions and AI guesswork. That's why AI-generated content often looks and feels the same (aka AI Slop). This n8n workflow grounds every decision - from creative optimisation to pre-testing and validation - in authentic human perspective. The result? AI-generated content that is genuinely differentiated + relevant for target audiences 🎯

Year on year comparison: Thinking about doing a startup vs actually doing a startup 😩