The Factory That Dreams: 39 AI Agents, No Framework - Rushabh Doshi, Machinecraft
https://www.youtube.com/watch?v=jtzh-GBXBWc
AI Engineer · 9 min · 155 sentences · sentence-level transcript from YouTube's caption track; each line begins at the time shown

[00:00] Okay, I want to tell you a story about a factory that taught itself how to remember.
[00:07] Hi, I'm Rushabh.
[00:08] I run Machinecraft, a 100 people factory in India.
[00:11] No data science team, no ML budget, none of that.
[00:15] And somehow we ended up building a 36 AI agent that runs our entire go-to-market.
[00:21] I think that's still a little ridiculous.
[00:24] Let me show you how it happened and why you can do the same thing.
[00:30] So, here's the thing about our company.
[00:32] From the outside it looks like machines and metal.
[00:34] But the actual company, the part that matters, isn't the machines, it's the knowledge.
[00:40] Who the customer is, what we quoted them in 2019, why that one machine needed that weird custom tweak.
[00:47] And for three generations, all of that lived in exactly three brains.
[00:51] Initially my grandfather's, then my father's, and now mine.
[00:57] Which is a genuinely terrifying way to run a company.
[01:00] When you sit with it.
[01:02] A lot of people have joined us, people have left us, the revolving door never stopped.
[01:08] And every single time someone walked out, a chunk of our brain walked out with them.
[01:15] We weren't scared of the competitors, we were scared of forgetting.
[01:18] Or waking up one day and realizing the whole company only existed inside two increasingly tired heads.
[01:27] So, I had an idea.
[01:29] I'll be honest.
[01:31] Sounded insane first.
[01:32] But what if instead of writing the knowledge down in some document nobody ever reads, what if we grew a brain that just held it?
[01:42] Not a chatbot you poke at, a twin of the company.
[01:46] I didn't hire a sales team.
[01:48] I tried to build one.
[01:50] A quick detour because you need to know how messy this is.
[01:55] We make thermoforming machines.
[01:57] They heat up a plastic sheet and shape it.
[02:01] Same core machine, but it ends up making hydroponic farm trays, spa bathtubs, EV car panels, medical casings, and even packaging.
[02:10] Seven totally different worlds, seven totally different buyers.
[02:14] So, this brain couldn't just memorize a brochure.
[02:18] It had to know which universe a given customer lives in.
[02:24] Step one was almost boringly simple.
[02:27] Feed it everything, and I mean everything.
[02:30] Years of quotes, drawings, payment schedules, timelines, email threads, hundreds of gigabytes of our own private history.
[02:38] Not the public internet, our internet.
[02:42] And here's the plot twist.
[02:44] The part that surprises every engineer I tell this to.
[02:48] We never trained a model.
[02:50] No GPUs humming in the basement, no fine-tuning.
[02:53] We just looked at all the history, chopped it into bite-size chunks, and let off-the-shelf models read it and pull out the facts.
[03:02] We stored the meaning of each chunk as vectors and relationships.
[03:07] Who's connected to what as a graph.
[03:10] The brain isn't a smarter model.
[03:12] It's actually a really, really well-organized memory.
[03:17] Now, this is where it gets a little weird in a good way.
[03:22] We stopped thinking of Era as a software and started thinking of it as something we were raising.
[03:27] So, we gave it a body modeled on biology.
[03:30] Senses to figure out who it's talking to, a gut to digest the documents into facts, a memory, a dream cycle, an immune system to fight off bad information.
[03:42] Why biology?
[03:43] Well, because evolution already spent a billion years solving how do you stay coherent over time?
[03:50] We just copied the homework.
[03:52] Okay, so the big question.
[03:54] Why 36 agents instead of one genius mega prompt?
[03:58] Because, and you already know this if you've ever tried it, one prompt that's supposed to do everything ends up doing everything badly.
[04:07] So, Ira isn't one mind, it's a pantheon.
[04:10] A whole cast of specialists.
[04:13] Each one has exactly one job.
[04:16] Athena runs the room.
[04:18] Prometheus owns the sale.
[04:21] Plutus does pricing.
[04:23] Hephaestus knows every machine spec cold.
[04:27] Vera fact checks everything, and Memnon, my favorite, guards corrections.
[04:33] So, the second a human fixes something, it stays fixed forever.
[04:38] One agent, one job.
[04:40] It's a team, not a hero.
[04:43] And here's the cool part.
[04:45] They hold meetings.
[04:46] Athena pulls in specialists.
[04:49] They actually argue, and a single answer comes out the other side.
[04:53] It's like having a board room that never sleeps, never gets tired, and somehow has no ego.
[05:00] So, what does all this actually run?
[05:03] Honestly, the whole front business.
[05:06] Everything between a stranger exists somewhere, and now they're a customer.
[05:13] Nine concrete jobs every single day.
[05:15] Outbound emails that actually reference my real world.
[05:19] Account briefs built from cross-checked truths before a call.
[05:24] Quotations.
[05:25] A swipe left, swipe right mode for outreach.
[05:29] Reviving dead leads, which I call blast from the past.
[05:34] Inbound replies, and figuring out before we waste an hour whether a company is even a fit.
[05:40] Nine jobs, one operator who never sleeps.
[05:45] Where does all this live?
[05:47] One cursor tab.
[05:48] That's genuinely it.
[05:50] You type and Eira reaches out with a dozen hands.
[05:53] Searches the knowledge base, reads the inbox, drafts the email, builds the code, and then shows you before anything actually goes out.
[06:03] Under the hood, it's genuinely a real stack.
[06:06] Not a demo held together with duct tape.
[06:08] Databases for vectors, for relationship graph, for the CRM.
[06:13] Three different model providers, each picked for the job it's actually best for.
[06:17] Tools for Google, for swallowing documents, for every communication channel, plus monitoring.
[06:23] So, we can see what it's thinking.
[06:26] All of it, every capability exposed as 213 tools over one protocol.
[06:32] And the golden rule, the one we never break, Eira drafts, human sends.
[06:38] Now, memory.
[06:40] And this is the part where most AI quietly lies to you, because a raw language model is basically a goldfish.
[06:46] Brilliant for about 30 seconds, and then you close the tab and forgets you ever existed.
[06:52] So, we engineered memory on purpose, in layers.
[06:55] Working memory for the last few minutes.
[06:58] Pinned facts about someone who who is.
[07:02] Episodes, whole conversations as little stories.
[07:05] Relationships with warmth that grows from stranger to trusted.
[07:10] And a bouncer at the door.
[07:12] A salience gate that decides what's even worth remembering, so the brain doesn't fill up with junk.
[07:19] When two facts disagree, corrections win.
[07:22] Continuity without making things up.
[07:26] And then, I genuinely love this part.
[07:28] At night, it dreams.
[07:32] Every night, Eira runs a sleep cycle.
[07:35] It replays the day, locks in useful stuff, hunts for contradictions, gently forgets the stale junk, and turns the day's work into reusable skills.
[07:45] In the morning, there's a little dream report waiting for me to read.
[07:49] Here's what I consolidated.
[07:51] Here's what I Here's what I let go of.
[07:54] Here's what I figured out while you were asleep.
[07:56] The thing literally gets smarter overnight.
[08:01] And here's the part I care about the most.
[08:03] Every agent has a conscience.
[08:05] And it is emphatically not to be helpful, be harmless.
[08:12] It's a soul file written from the principles of a Jain family business that's been doing this for the last three generations.
[08:20] Five old ideas turn into engineering rules.
[08:23] No single source has the whole truth.
[08:25] So, cross-check before you speak.
[08:28] Never say things absolutely.
[08:30] Cite the document and the date.
[08:33] Do your own job, not someone else's.
[08:35] Report the truth even when the truth is ugly.
[08:38] And nobody works alone.
[08:40] Ancient philosophy running as guardrails in production.
[08:44] Now, let's talk money.
[08:46] Because this is the part that should make the whole industry a little uncomfortable.
[08:50] There was no training bill.
[08:51] Zero.
[08:53] The expensive part was never compute.
[08:54] It was teaching a company to remember itself.
[08:57] An agency quoted us 230 grand to build this.
[09:01] We built it for around 30.
[09:03] That's cheaper than a nice watch.
[09:05] And it runs on a couple of thousand dollars a month.
[09:09] So, here's the move.
[09:11] We pulled the whole architecture out and made it forkable.
[09:14] We call it Brain OS.
[09:16] It ships as an empty nervous system.
[09:18] The agents, the memory, the dream cycle, the soul file.
[09:22] All there, completely blank.
[09:24] You pour your own company's truth into it and from inside out.
[09:28] Because here's the thing nobody can outsource for you.
[09:32] Only you can build your company's brain.
[09:37] We are a 100 people factory with no data scientists.
[09:40] If we can grow a brain, you can too.
[09:42] We are not selling ours to you.
[09:44] We are helping you build your own.
[09:46] forkmybrain.org Go build something that remembers.
[09:51] Thank you.