Skip to content

By playing, you agree to uListen's Terms and Privacy Policy and YouTube's Terms of Service.

Sequoia CapitalSequoia Capital

Factory's Matan Grinberg: The Coming ‘Dark Factory’ Where Software Builds Itself

Factory started building fully autonomous coding agents in April 2023, two years before enterprises were ready. Matan Grinberg now says this is indistinguishable from being wrong.

Matan GrinbergguestPat GradyhostSonya Huanghost
Jul 21, 202651mWatch on YouTube

Summaries, chapters, transcripts, and FAQs on this page are made by uListen, not YouTube.

At a glance

WHAT IT’S REALLY ABOUT

Factory’s vision: model-independent droids powering a “dark” software factory

  1. Matan Grinberg said Factory’s differentiation for enterprises is model independence, driven by widespread fear of vendor lock-in and volatility among model providers.
  2. He described Factory’s early struggle as being “two years too early,” including a period where the team refunded customers because the product wasn’t creating developers who were “obsessed” with it.
  3. He argued that adoption changes came more from developer behavior and workflow readiness than purely from model capability improvements, with the Droid CLI providing a key “meet developers where they are” interface.
  4. He outlined how “token maxing” is evolving into cost rationalization, with Factory’s router dynamically selecting models (including open models) based on task needs and organizational constraints.
  5. He framed the end-state as a “software factory” and ultimately a “dark factory” where most tokens are spent asynchronously by autonomous agents, creating tighter feedback loops on where dollars and tokens should go.

IDEAS WORTH REMEMBERING

5 ideas

Factory optimizes for “creating obsessed customers,” not performing customer-obsession rituals.

He argued that “customer obsession” is an input metric and can be misleading; the output metric that matters is whether customers become obsessed with the product. This framing drove decisions like refunding revenue when Factory felt it wasn’t delivering an experience developers loved.

Enterprise demand is shifting toward model independence and away from single-vendor dependence.

He said enterprises have “scars” from cloud-era pricing and migration pain, and they now resist putting their fate in any single model provider’s hands. Factory positions itself around modularity and the ability to hot-swap models as better/cheaper options emerge.

Factory chose trust over short-term revenue by refunding customers when the product wasn’t meeting the bar.

When Factory realized it had signed contracts while “the product was not good,” he said they proactively refunded customers rather than risk churn and broken trust. He described it as one of the hardest moments, but central to maintaining credibility for when the product would be ready.

The breakthrough was changing the interaction pattern (Droid CLI), not just waiting for better models.

He said the early approach required too big a behavioral leap—fully autonomous agents before developers were even comfortable with tools like autocomplete—making the product “two years too early.” He pointed to the Droid CLI launch (September 26, 2025) as a shift to meeting developers where they work.

Supporting many models can produce a better harness than co-designing harness + single model.

He claimed a multi-model harness improves performance by avoiding overfitting to one model’s quirks, analogizing that models are to a harness what data is to a model. He illustrated with differing tendencies of Opus vs GPT 5.6 and the need for consistent user experience (e.g., reliably maintaining to-do lists across models and long-session compactions).

WORDS WORTH SAVING

5 quotes

Like, our job is to build something so good that our customers themselves become obsessed with us.

— Matan Grinberg

Thank you, but being two, three, two or three years early is the same as being wrong.

— Matan Grinberg

We realized this and we proactively gave all of those customers their money back.

— Matan Grinberg

Much to the chagrin of many of my friends at OpenAI and Anthropic, this is not true. If you build a harness that supports different models, that harness will be better.

— Matan Grinberg

This idea of like a dark factory where like the lights are off and things are just happening, that is where software development's going.

— Matan Grinberg

Read the available transcript
Creating “obsessed customers” vs input metricsEnterprise vendor lock-in fearsModel independence and modularityRefunding revenue to preserve trustDroid CLI launch and workflow fitHarness performance: caching, compaction, tool useMulti-model harness vs model+harness co-design claims about labs and performance differences across models (Opus vs GPT 5.6)

SPEAKERS

  • Matan Grinberg

    guest

    Co-founder and CEO of Factory, building autonomous software-engineering agents (“droids”).

  • Pat Grady

    host

    Sequoia Capital partner and podcast host/interviewer.

  • Sonya Huang

    host

    Sequoia Capital investor and podcast host/interviewer.

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

Read along on your phone, search it beside the video, or ask your favorite AI app what was said.