One context for code, collaboration, and AI. vx: data-lake
Joined April 2023
- Tweets331
- Following410
- Followers86
- Likes662
Git’s deepest innovation is not just storing commits — it is the workflow around change: branches for parallel exploration, history for traceability, issues and PRs for human collaboration, and a distributed model for global contribution.
New on our Engineering blog: Creating in-house source control.
In May, we merged around 2,500 code changes per day. In September, we merged more than 30,000 per day. As a result, we became bottlenecked by developer infrastructure built for human (rather than agent) scale.
Here, we share how we built a new version control system—and fully migrated our monorepo from git—in five months: anthropic.com/engineering/re…
Data is the center of the universe; agents are ephemeral.
In the Agent era:
Software as fixed SaaS becomes redundant.
Agent dynamically synthesizes software capabilities on the fly.
Data & Context remain the only core, enduring asset.
I've actually stopped using the word "harness." I now think in terms of two things: agent runtimes and agent applications.
Agent runtimes should be very small. Their core job is to connect an LLM API to an execution environment. That environment could be a full OS or a language VM, like a JavaScript runtime. Almost all that matters here is robustness. An agent runtime must be able to keep an agent running forever. Literally forever.
Agent applications are everything else. Yes, literally every piece of software in the world.
If software isn't built for agents, they use it through computer use or browser use. These capabilities can be built into the agent runtime or provided by another agent application.
If software is agent-native, agents should be able to use it directly. The experience of an agent using that software—what we call "agent experience"—needs to be carefully designed, at least for now, to make efficient use of both time and tokens.
Another key piece is agent identity. Agent applications should recognize each agent individually, rather than just the browser or computer it's using. This allows each agent's actions to be recorded in detail and audited in real time or after the fact.
Everything we're building at @botiverse follows from this view.
As personal agents become the interface between consumers and companies, this kind of product provenance may become a key signal for agents to evaluate quality.
Could there be a “Wikipedia for products” built on Git?
Not just a page describing what a product is today, but a living history of how it evolved — versions, iterations, design decisions, user feedback, and improvements.
Git doesn’t stop at code.
As AI starts working with data, models, workflows, and other digital assets, the same primitives become increasingly important.
If you’re working at the intersection of Git, Data, and AI, this is worth reading.
As a team building at the intersection of Git + Data + AI, we have to recommend this article.
It’s one of the most worthwhile Git guides we’ve come across — not because everyone needs to become a Git expert, but because understanding Git changes how you think about building.
Git changed software development by making experimentation cheap.
Git-like data infrastructure could do the same for Agentic orgs：
1. Downstream consumers never see corrupted data.
2. And nobody debugs failures under pressure.
The key idea is not just versioning data.
It’s giving Agents a safe, reproducible environment to experiment against real data — without copying massive datasets every time.
Metadata → zero-copy → delta → full copy
What if Agents could branch data and run experiments against it? @sspaeti explains here:
ssp.sh/blog/git-for-data-the…
Branch the data
→ Let Agents experiment against it
→ Test on real data
→ Compare results
→ Deploy what works
→ Roll back when needed
The key idea: don’t copy petabytes of data just to create a branch.
Separate metadata from data, and you can get Git-like branching, snapshots, rollback, and isolated testing with zero-copy or minimal data movement.
Want to understand how Git For Data works in practice?
Check out this deep dive by @sspaeti comparing Git-like tools that separate metadata from data:
ssp.sh/blog/git-for-data-too…
It covers:
1. LakeFS
2. Dolt
3. Nessie
4. MotherDuck
5. Neon, Supabase, Bauplan & more
no GitDataAI😂
Branching, isolation, reproducibility and merge are becoming fundamental primitives for data infrastructure — and even more important as AI agents begin to work directly with data.
This is exactly the direction we’re exploring with Git+Data+AI.
Thanks Simon for mapping out！
GitData.AI retweeted
Engineers care about two things in production, says Ciro Greco from Bauplan: downstream consumers never see corrupted data, and nobody debugs failures under pressure.
His answer is git for data. No job writes to production tables directly.
ALT Git for Data
Once every data change can happen in an isolated branch, failure is no longer something you have to debug against production. The failed state itself becomes an asset you can inspect, reproduce, learn from, and improve.
And as AI agents increasingly work with data, this becomes even more important. Agents need a versioned and legible data foundation they can safely experiment against — not just a production database they can read from.
Thanks for mapping out the growing ecosystem. This is exactly the direction we believe the data infrastructure is moving toward.
The deeper shift is making data assets behave more like code: branchable, reproducible, inspectable, mergeable, and collaborative.