The Codebase Is No Longer a Moat

John Jeong — Tue, 07 Apr 2026 00:59:18 +0000

The codebase is no longer a moat

For the last two decades, the software industry believed a simple
rule: keep your code closed and competitors cannot catch you. A rule
built on fear, and on a world where writing software required armies
of engineers and thousands of hours.

AI has rewritten those economics. A small team with strong taste now
ships at a pace that used to require a company of fifty. Tools like
Cursor, Devin, Zed, Claude Code, GitButler, and Warp generate
features in hours instead of weeks — and can refactor entire systems
in a single morning.

The real advantages now are taste, execution, and trust.

Why open source is having a second wave

Open source dominated infrastructure for years — Linux, PostgreSQL,
Docker, Kubernetes, Redis, Python, Rust. Applications never followed
the same pattern. Most teams assumed real product companies needed
proprietary edges, branding control, or complex business models.

OBS Studio, Blender, VLC, Krita — people used these apps every day
without thinking about whether they were open source. They used them
because they worked well, respected their users, and did what they
promised.

Open source is now moving past infrastructure into applications:
productivity tools, creative software, knowledge systems, research
platforms, AI interfaces, browsers.

AI is accelerating open source, not replacing it

AI reduces the cost of contribution and speeds up iteration. It
expands what a small team can accomplish and lowers the barrier to
building polished interfaces and high-quality UX.

Open source no longer means slow, clunky, or "good enough for free."
With AI, open source can match or exceed the pace of any closed team.
Speed combined with transparency and a passionate community creates
something more durable than secrecy.

The real moat has shifted

When code becomes cheap to generate, product philosophy becomes the
moat.

Can you build something that works well, respects its users, and earns
enough trust that people depend on it?

Closed code used to signal ownership. Now it mostly signals fear.
Open source signals confidence — users own their tools rather than
being locked inside them.

In a world shaped by opaque AI systems, transparency has real value.
In a world where data privacy matters more each year, user control
is essential.

Why I built OSS Friends

I wanted to give back to the community that shaped how I build.

OSS Friends is a directory of
open source applications that inspire me — not infrastructure or
libraries, but actual apps people use every day. Tools that prove
open source is a product philosophy, not just technical infrastructure.

I want to highlight teams pushing software toward transparency, user
ownership, and good taste. If you know an open source app that
deserves attention, drop it in the comments.

The future is open, not because of ideology, but because of reality

The shift is practical, not political.

Companies that understand this early will build the next decade of
software. Those clinging to secrecy will find they cannot keep pace
with a global open source community powered by AI.

The moat has moved.

John Jeong is the Cofounder of Char — the open source Granola AI alternative.

Stop giving away your data for free after meetings

John Jeong — Thu, 24 Apr 2025 04:17:29 +0000

1. “AI in the room” means a copy of your meeting in the cloud

Most cloud‑based notetakers work the same way:

Record & upload – the app (or its meeting‑bot) streams audio/video to the vendor’s servers.
Process & store – speech‑to‑text and large‑language‑model (LLM) summarization happens in the vendor’s cloud. The raw recording and the transcript usually stay there.
Reuse & share – vendors may train models on “de‑identified” data, pass content to third‑party processors, or keep it indefinitely for “service improvement.”

Even if data is encrypted in transit, once it rests on someone else’s servers you inherit every one of their risks—breaches, subpoenas, jurisdictional conflicts, and policy changes you can’t control.

2. Cloud‑AI risks in the real world

Risk	What it looks like in practice
Model training on your words	Otter.ai explicitly says it trains its speech models on your de‑identified recordings and transcripts.
Indefinite retention	Fathom retains meeting recordings “until you delete them,” i.e., forever by default.
Third‑party exposure	Fireflies routes content through external STT & LLM vendors (they say those vendors have “zero‑day retention,” but it’s still outside your perimeter).
Consent & compliance failures	UMass banned Otter.ai after finding it violated the state’s all‑party consent law.
Accidental leaks	An Otter transcription glitch in 2024 shared a confidential post‑meeting chat with unintended recipients.

For teams in finance, healthcare, legal, or government, any one of these can break HIPAA, GDPR, FINRA, or internal NDA requirements.

3. Competitor snapshot

Otter.ai – Trains models on user data; allows staff access with consent; stores content on U.S. servers.
Fireflies.ai – Claims “no training,” but still funnels calls through cloud vendors and integrates with CRMs, increasing the blast radius.
Fathom – Keeps recordings indefinitely; shares device/usage metadata with partners, which can create traceable patterns.
Granola – Doesn’t store raw audio, but does store transcripts on its servers for sharing.

All four products ultimately rely on a cloud repository you don’t control.

4. Hyprnote’s local‑first guarantee

On‑device speech‑to‑text – Hyprnote ships with Whisper.cpp, so your audio is transcribed on your Mac.
On‑device summarization – The LLM that cleans up your notes also runs locally, never leaving the machine.
No bots, no streaming – Hyprnote sits in the menubar and listens only on your machine—no meeting invites, no hidden guests.
Full user control – Delete a note, and the data is gone. Unplug from the internet, and everything still works.
Open‑sourced core – Anyone can audit the code to verify that we’re not phoning home.

Result: your compliance officer can sleep at night, and your users keep the convenience of AI without surrendering the contents of every call.

5. Why local‑first matters for strict‑compliance teams

Regulatory alignment – Keeping PHI or customer PII off external servers avoids HIPAA Business Associate headaches and EU data‑transfer clauses.
Breach surface reduced – Attackers would need physical access to the user’s machine to get the raw audio.
Jurisdiction clarity – Your data lives under the same legal regime as the device it’s stored on—no surprise subpoenas from another country.
True opt‑in sharing – You decide if and when to sync or share; Hyprnote never assumes consent.

6. Take back control of your meetings

Cloud notetakers have made life easier—but at the price of handing your most candid conversations to someone else’s servers. Hyprnote proves you can have AI convenience and privacy.

Ready to keep your notes where they belong? Download Hyprnote and try local‑first note‑taking today.

Try Hyprnote for free forever, safely

DEV Community: John Jeong

The Codebase Is No Longer a Moat