--- title: Why Factory? company: Factory (factory.ai) role_target: Commercial / Strategy / GTM created: 2026-06-04 status: prep --- # Why Factory? **The spine of the answer (Darsh's thesis):** Tokens are becoming an enterprise budgeting nightmare. The answer is model routing — not everything needs frontier intelligence — and whoever owns the *primary surface* owns the routing decision, the usage data, and the budget. Factory is the cleanest bet on that, and the labs structurally can't build it themselves. --- ## The answer (full, ~written/long-form version) I'll skip the mission paragraph and give you the strategic reason, because it's the one I actually believe. The most underpriced problem in enterprise AI right now is that **tokens are a budgeting nightmare.** Seat-based SaaS trained every CFO to forecast software as a clean per-head line item. Consumption pricing breaks that — spend scales with usage, not headcount, and it's nearly impossible to cap or forecast before the invoice lands. The first wave of adoption hid this because volumes were small. The second won't. The moment agent usage scales, finance starts asking the question I've spent my career on the other side of: what is this spend, and what does it return? Factory is the cleanest bet on the answer. **Not everything needs frontier intelligence.** Most engineering work — the lint, the rename, the search, the routine PR — doesn't need the most expensive model in the world; it needs the cheapest model that clears the bar. Routing is how you turn an unbounded token bill into something a finance team can actually govern, and Factory is building the routing and cost layer as a first-class enterprise surface, not a bolt-on feature. That maps directly onto my edge: I come from strategic finance, I think about value capture for a living, and I build agents myself — I already route my own work, small models for the cheap passes, frontier models for the hard reasoning. The instinct Factory is productizing is one I use every day. But the deeper reason is **gravity, not cost.** Whoever owns the primary surface — the place where the developer actually issues intent — owns the routing decision, sees the usage, and captures the budget. The model underneath becomes swappable; the surface does not. I wrote a piece a while back about Uber and the risk of the OS becoming the front door: *be careful who owns intent, because you can't price what you no longer control access to.* Factory is that lesson applied to software development. Own the surface, and you own the routing layer — and the routing layer is where the value pools as the models commoditize beneath it. And here's what makes me think the moat is real rather than rented: **the labs structurally can't build this.** Anthropic and OpenAI are incentivized to maximize consumption of their own frontier tokens — they will never ship the honest router that sends your easy tasks to a cheaper model or a competitor. A neutral, cost-optimizing control plane has to come from someone who doesn't sell the tokens. That's the gap Factory sits in, and it's why "the model providers will just add this" is weaker than it sounds on the exact dimension a CFO cares about most. So that's why Factory, why now, and why me: the commercial problem here — pricing, governing, and defending the value capture of autonomous agents in budget-conscious enterprises — sits precisely at the intersection of the three things I'm actually good at. Finance, monetization, and building. I'm not coming to admire the Droids. I want to help own the layer where the spend becomes governable and the gravity compounds. --- ## The 60-second spoken version Honestly? The thing that pulls me to Factory is that tokens are turning into a budgeting nightmare for enterprises, and almost nobody is solving the *finance* side of that. Seat pricing was predictable; consumption pricing isn't, and the moment agent usage scales, finance starts asking what the spend is and what it returns. Factory's bet is the right one — not everything needs frontier intelligence, so you route, and you build the cost layer as a real surface. And the part I find genuinely defensible is that the labs *can't* build this honestly — they're incentivized to sell you more of their own frontier tokens, not route you to something cheaper. A neutral routing layer has to come from someone who doesn't sell tokens. That's Factory. And it sits exactly where I'm strongest: I'm a finance and strategy person who also builds and routes agents myself. I'd want to own the commercial side of making that spend governable. --- ## Cold email — v3: the one to actually send (to Troy, role-specific) ⭐ *Operative version. A posted role (Strategic Finance Associate) + a named hiring manager (Troy, likely the finance lead) + your single strongest asset: you already do this exact job for a consumption-model agentic product. Lead with the credential; hold the routing/ROI thesis for the call. Short beats clever here.* **Subject:** Strategic Finance Associate — flagging my application Hi Troy — Saw Factory is hiring a Strategic Finance Associate, and wanted to flag my application directly. I lead Strategic Finance for the agentic development products on Unreal Engine (Epic's game-dev platform) — so I spend my days (and a fair few nights) on the exact problem Factory is built around: the consumption-model economics of agentic software. I rarely see a role map this cleanly onto what I already do. Any chance you'd have 20 minutes? Best, Darsh Shah LinkedIn · GitHub · Website · 224.706.4188 *Optional +1 line, only if you want a sliver of POV before the call:* "Happy to share how I'd think about pricing predictability as the lever that turns self-serve usage into enterprise commits." **Edits from your draft:** tied your work explicitly to Factory's business ("the exact problem Factory is built around") so it reads as *I get why this matters here*, not just *I'm qualified*; swapped the generic "would love to chat" for specific enthusiasm ("I rarely see a role map this cleanly"); cleaned the signature. Kept it ~75 words — the brevity is the strength. --- ## Cold email **To:** Matan Grinberg, CEO & co-founder. For a strategy/commercial role at this stage, the founder is the right cold-email target — the argument below is a CEO-level thesis about value capture, not a sales-team note. Likely address `matan@factory.ai` — **verify before sending** (LinkedIn/X/a mutual). Retarget to whoever leads Revenue/GTM only if you'd rather go that route. **Subject:** The routing layer is the moat the labs can't build *(alts: "Tokens are a budgeting nightmare — and the labs can't fix that" · "The CFO problem hiding inside Factory's thesis")* --- Matan — Most enterprise-AI pitches are still about who writes better code. I think the next wave is duller and bigger: tokens are becoming a budgeting nightmare, and finance is about to start asking what the spend returns. Seat pricing was predictable; consumption isn't. Factory is the cleanest bet on the answer — not everything needs frontier intelligence, so you route, and the cost layer is as much a finance product as an engineering one. The part I find genuinely defensible: the labs can't build this honestly. They're paid to maximize their own frontier tokens; they'll never ship the router that sends easy work somewhere cheaper. A neutral routing layer has to come from someone who doesn't sell tokens. That's you. I'm a strategic-finance operator who also builds and routes agents daily — I've been the person who has to justify the AI spend and own the number. That's the exact commercial problem you're solving inside buyers like Morgan Stanley and J.P. Morgan. Worth 15 minutes? I'll bring a point of view on where the value pools, not a résumé. Darsh darshshah.xyz --- **Ultra-short variant (~90 words, lower friction):** > Subject: The moat the labs can't build > > Matan — a thesis, not a pitch: tokens are becoming an enterprise budgeting nightmare, and the labs can't fix it, because they're paid to sell their own frontier tokens and will never ship an honest router. A neutral routing/cost layer has to come from someone who doesn't sell tokens. That's Factory. I'm a strategic-finance operator who builds and routes agents daily, and I've been the one who has to justify the AI spend. Worth 15 minutes on where the value pools? > > Darsh · darshshah.xyz **Before you send:** - **Verify the address** (likely `matan@factory.ai`). A bounce is the only unforced error here. - **One link, no attachment.** Use `darshshah.xyz`, or point straight at *The Repricing of Enterprise Software* — it's literally this thesis. Offer the résumé only if he bites. - **Timing:** Tue–Thu morning. One follow-up after ~5 business days, then stop. - **Don't soften "That's you."** Confidence about *his* business is what separates this from a templated applicant — it's the whole reason a founder replies. --- ## Cold email — alt: ROI / measurement lead (recommended) *Same target (Matan Grinberg). This version leads with the ROI-measurement thesis instead of routing — less commoditized, more squarely the finance lane, and it folds in the dual-motion insight (ROI is the hinge that converts self-serve usage into enterprise expansion). Use this as the primary; keep the routing version above as a B-test.* **Subject:** Enterprise AI coding gets decided at the ROI review *(alts: "The CFO is the real bottleneck, not the model" · "Whoever makes the CFO comfortable wins enterprise")* --- Matan — Most of the field is still racing on capability — who writes better code, finishes more tasks. I think the enterprise winner gets decided somewhere duller: at the ROI review. Every AI-coding deployment I've watched stalls on the same question — prove the productivity gain — and the usual answers (lines of code, DORA, "devs feel faster") don't survive a budget meeting. Whoever makes the CFO comfortable unlocks the budget expansion. That's why Factory Analytics is the most interesting part of your stack to me — tying agent usage to outcomes is the real unlock, not the model. It's also the hinge of your funnel: credible ROI is what turns a self-serve team's rising token bill into a committed enterprise deal instead of churn. Proving and pricing the value of autonomous agents to a finance org is my lane. I'm a strategic-finance operator who also builds and routes agents daily — I've been the person who has to justify the spend and own the number. Worth 15 minutes? I'll bring a view on how to make that ROI case credible, not a résumé. Darsh darshshah.xyz --- **Why this one is the stronger lead:** routing is contestable (OpenRouter/Portkey exist; the labs do intra-family routing). "Credible ROI measurement is the enterprise unlock" is harder to argue with, it's the actual bottleneck every deployment hits, and it's the purest expression of your finance edge. Routing then becomes *supporting color* inside this story, not the headline. --- ## Talking points (internalize these, don't recite) - **The wedge is finance, not capability.** First wave of enterprise AI = "who writes better code." Next wave = governance, reliability, and *cost*. The token bill is the unsolved CFO problem. Lead with that — it's your lane and most candidates won't see it. - **Routing = turning an unbounded bill into a governable one.** "Not everything needs frontier intelligence" is the economic core. ~Most agent tasks are routine; routing them to cheap models collapses cost without hurting quality. - **Surface = gravity (your own "who owns intent" thesis).** Own where intent is issued → own routing → own usage data → own the budget. Model is swappable; surface is sticky. Tie explicitly to your Uber/leverage post. - **The labs' conflict of interest is the durable moat.** Anthropic/OpenAI won't route you *away* from their frontier tokens. Neutral routing must be third-party. This is the rebuttal to "model providers move down." - **You're the buyer they're selling to.** Their customers (Morgan Stanley, J.P. Morgan, EY) need to "explain to finance why AI spend is rising." That sentence is your entire background — you've sat on the side of the table that justifies the spend and owns the number. - **You're also the user.** You build agents (Cadence, your own writing/stylometry tooling) and already route models by task. You're the "developer fluent with agents" Grinberg says will outpace everyone — and you can speak product *and* finance. ## Honest risks (raise one yourself — it signals you're not a sycophant) - **Differentiation collapse / the squeeze.** Code hosts (GitHub, GitLab, Atlassian) move *up* into agent governance; model labs move *down* into admin controls. Factory is squeezed between the workflow owners and the intelligence owners. Your counter: the *routing/cost* dimension is where the squeeze is weakest, because the labs are conflicted and the hosts are tied to their own ecosystems — but the control plane only survives if it keeps earning its place with real workflow value. - **"Model-switching" isn't unique.** Grinberg has pitched model-agnosticism as the differentiator; press (WSJ/TechCrunch) note Cursor isn't single-model either. So don't lead with "they switch models" — lead with *cost-aware routing as a finance/governance product*, which is the sharper, less-commoditized version. ## Factory facts (for grounding — verified 2026-06-04) - **Founded** 2023 by **Matan Grinberg** (UC Berkeley physics PhD dropout, CEO) + **Eno Reyes** (ML engineer). Origin: Grinberg cold-emailed Sequoia's Shaun Maguire (ex-physicist), who told him to drop out. - **Product:** "Droids" — autonomous, **model-agnostic, interface-agnostic** agents (CLI, IDE, Slack, browser) across the full SDLC. #1 on **Terminal Bench** at Series B launch. - **Funding:** Series A $15M (Sequoia, ~$120M val) → **Series B Sept 2025, $50M at $300M** (NEA, Sequoia, **Nvidia, J.P. Morgan**; angels Slootman, Arora, Levie) → **Series C Apr 2026, $150M at $1.5B** (Khosla led by **Keith Rabois**, + Sequoia, Insight, Blackstone). Unicorn. ~$220M raised total. - **Traction:** revenue reportedly doubling ~monthly for 6 months pre–Series B. - **Customers:** Morgan Stanley, J.P. Morgan, EY, Palo Alto Networks, Nvidia, Adobe, Adyen, MongoDB, Bayer, Zapier. **Heavy financial-services / regulated-enterprise lean.** - **Positioning:** **enterprise-first** (vs Cursor's developer-first DNA); a **control plane "above the models."** NEA thesis: *"The great opportunity is not beating the latest benchmark — it is building the enterprise abstraction which enables modernization of codebases."* - **Wedge:** **brownfield / legacy modernization** (COBOL, Fortran, Java 7); built a dedicated legacy-SWE benchmark. Differentiators framed as **governance, security, ROI measurement (Factory Analytics), and cost routing** — "the messy bits," not raw model quality. - **Competitors:** Cursor (developer-first IDE), Cognition/Devin (autonomous engineer), Anthropic Claude Code (peer/competitor — and note Factory sits one layer *above* the labs). ## Sources - [Factory raises $50M Series B — SiliconANGLE](https://siliconangle.com/2025/09/25/factory-unleashes-droids-software-agents-50m-fresh-funding/) - [Series B announcement — BusinessWire](https://www.businesswire.com/news/home/20250925993478/en/Factory-Unleashes-the-Droids-Raises-$50-Million-Series-B-from-NEA-Sequoia-Capital-NVIDIA-and-J.P.-Morgan) - [Factory hits $1.5B valuation to build AI coding for enterprises — TechCrunch](https://techcrunch.com/2026/04/16/factory-hits-1-5b-valuation-to-build-ai-coding-for-enterprises/) - [Series C $150M / Khosla — TechFundingNews](https://techfundingnews.com/factory-ai-150m-series-c-unicorn-ai-agents-developers/) - [Factory: The Platform for Agent-Native Development — NEA](https://www.nea.com/blog/factory-the-platform-for-agent-native-development) - [Factory AI brings cost routing to enterprise coding agents — Startup Fortune](https://startupfortune.com/factory-ai-brings-cost-routing-to-enterprise-coding-agents/) - [Factory Enterprise](https://factory.ai/enterprise)