←
Richard Feldman2,625 posts
Richard Feldman avatar

Richard Feldman

@rtfeldman

Let’s go with the ambitious approach.

◷ Joined September 2008▧ 117 Photos and videos
Evan Czaplicki avatar
⟳ Richard Feldman retweeted
Evan Czaplicki@evancz·8m

Tony Hoare on corporate languages and the price of reliability: "The price of reliability is the pursuit of the utmost simplicity. It is a price which the very rich find most hard to pay."

Post attachment
RWX avatar
⟳ Richard Feldman retweeted
RWX@rwx·23m

Brilliant moved CI from GitHub Actions to RWX and builds went from 20–25 min on GHA to ≃3 min on RWX. That’s an 85% reduction in CI/CD execution time. How it happened ↓

Post attachment
Isaac Van Doren avatar
⟳ Richard Feldman retweeted
Isaac Van Doren@isaacvando·Nov 16

Software Should Work will return on July 15-16 2027. Save the date, tell your friends, and stay tuned for the speaker lineup! @ssw_conf

Post attachment
Zed avatar
⟳ Richard Feldman retweeted
Zed@zeddotdev·Nov 14

“Agent archaeology”: piecing together why a teammate and their agent wrote code a certain way from scattered commits, transcripts, and conversations. Richard gave a talk at AI Engineer NYC last month about how we use Delta to avoid it. youtube.com/watch?v=Enn9dVrR…

Post attachment
Zed avatar
⟳ Richard Feldman retweeted
Zed@zeddotdev·Nov 13

Soon, your agents will be able to trace code back to the threads that produced it, giving them access to the decisions and reasoning the code alone can’t explain.

0:13
Hillel avatar
⟳ Richard Feldman retweeted
Hillel@hillelogram·Nov 12

There is an inescapable tradeoff between what a software technology can do (its abilities) and what you know for certain about what it does (its guarantees). For example: - Unicode is able to represent strings like `中文🌴`, but ASCII guarantees that all visible characters are exactly one byte. - Safe rust guarantees an absence of (certain kinds of) memory errors, but it isn't able to encode as many valid programs as unsafe rust can. - `map` can't do every computation a `foreach` loop can, but `map` always guarantees the output list is the same size as the input list. - Graphs can represent more kinds of data than DAGs, which represent more than trees, which represent more than lists. But DAGs are orderable, trees have at most one path between nodes, and lists guarantee one order. The tradeoff exists whether we're talking formats, data structures, languages, or tools. It's so fundamental because it's universal. Give me any software topic and I will give you an example of this tradeoff. What's causing this? There's a surprisingly simple mathematical explanation! Abilities correspond to `some` quantifiers over a set, while guarantees correspond to `all` quantifiers. `some` is "monotonically true" over the size of the set, meaning when we grow the set, it can flip from false to true but not from true to false. Conversely, `all` is monotonically false, meaning it can flip from true to false but not vice versa. In other words: to increase a technology's guarantees we need to "shrink the corresponding set", which by definition decreases its abilities. The ability-guarantee tradeoff is one of the running motifs of my book Logic for Programmers: not the subject of a chapter, but an idea that runs through many chapters. The tradeoff connects diverse topics like testing, optimization problems, and domain-driven design. If this interests you, the book is available in pdf, epub, and print formats. Check the link below! logicforprogrammers.com/ (Or, if you want to participate in the November charity drive, check out the link in the next Tweet!)

Evan Czaplicki avatar
⟳ Richard Feldman retweeted
Evan Czaplicki@evancz·Nov 10

Me when I want to avoid unexpected JSON so much that I write a typed functional query language. "Tony Hoare told me to do it in 1973!"

Post attachment
Evan Czaplicki avatar
⟳ Richard Feldman retweeted
Evan Czaplicki@evancz·Nov 9

Dijkstra on testing vs language design: "For what program structures can we give correctness proofs without undue labor?" With @elmlang you get: - no null pointers - no unhandled exceptions - no aliasing bugs - safety from supply chain attacks AND it's more fun than JS/TS 😅

Post attachment
Zed avatar
⟳ Richard Feldman retweeted
Zed@zeddotdev·Nov 5

Discord in GPUI! LFG! 🎉

On the left is the Discord you use daily. And on the right is a new version we're working on, rebuilt in Rust

Daniel Lemire avatar
⟳ Richard Feldman retweeted
Daniel Lemire@lemire·Nov 1

A CVE (Common Vulnerabilities and Exposures) is a publicly disclosed security flaw. They increased following a slight upward slope all the way to early 2024, when there was a sudden small increase... then starting in 2026, we went full exponential. In most instances, the vulnerabilities were pre-existing, so much of the increase is just our better ability to find the faults. This is of historical importance. We have never witnessed anything of the sort.

Post attachment
Antonio Scandurra avatar
⟳ Richard Feldman retweeted
Antonio Scandurra@as__cii·Oct 29

I don't think version control for agents will be solved by making Git backends faster or forcing collaboration onto one machine in the cloud. It needs a more ambitious primitive. We've spent the last 12 months building one. So excited to share how DeltaDB works under the hood!

Zed@zeddotdevOct 29

DeltaDB is a new, Git-compatible version control system for agent context. It keeps every edit linked to the conversation that produced it. @as__cii takes a deep dive into how we built it with Rust, SQLite, and Durable Objects: delta.dev/blog/fine-grained-…

Zed avatar
⟳ Richard Feldman retweeted
Zed@zeddotdev·Oct 29

DeltaDB is a new, Git-compatible version control system for agent context. It keeps every edit linked to the conversation that produced it. @as__cii takes a deep dive into how we built it with Rust, SQLite, and Durable Objects: delta.dev/blog/fine-grained-…