Creator of @elixirlang. Chief Adoption Officer at @dashbit, where we build tidewave.ai and livebook.dev.
Joined November 2007
- Tweets 2,398
- Following 93
- Followers 55,106
- Likes 1,919
157 Photos and videos
Pinned Tweet
11 Nov 2025
It is finally here: Tidewave now supports Claude Code and OpenAI Codex.
Tidewave unlocks the full-stack potential of your favorite coding agent by tightly integrating it with your web app and web framework at every layer, from UI to database. More info 👇
14
48
320
87,144
José Valim retweeted
In my setup, I gradually tune the prompts to encourage agents to use ExAST whenever possible
github.com/elixir-vibe/ex_as…
2
1
12
412
I wanted Codex to find usages of [key: foo | bar] in Elixir's codebase. It grepped first, but got too many false positives.
BUT, since Elixir's AST is quite uniform, it chose on its own to write a script that parses all files, scans the AST, and gave me precise locations.
WILD.
6
8
102
4,783
José Valim retweeted
Phoenix LiveView 1.2 is released with colocated css and more! Check the announcement for details!
phoenixframework.org/blog/ph…
2
29
134
7,236
José Valim retweeted
Jun 10
NEW: malware developers added nuclear & biological weapons text to to their spyware.
Goal? To trigger LLM safety refusals... so that their spyware wouldn't be analyzed by an AI security scanner.
Cleanest practical example I can think of for why over-indexing on first order safety alignment is risky.
When closed (and open) models ship with aggressive refusals, they will be sprinkled with second-order blindspots that attackers will discover...and exploit.
We are only in the earliest days of attackers leveraging these features, and it wouldn't surprise me if users systems that need to handle complex cybersecurity issues demand that models be less safety-blunted.
In the weeds: @SocketSecurity's post also shows why intention matters in how you design a malware analysis pipeline to avoid prompt manipulation.
H/T to colleagues that shared this with me socket.dev/blog/mini-shai-hu…
226
2,153
12,635
1,543,778
Jun 12
Wholeheartedly agree.
Also, after talking to people who have been "running loops", it is very clear that setting up the loop is a lot of work and takes tons of time, which almost no one is talking about.
If you automate the wrong things or in the wrong ways, they will crumble.
Jun 12
If we want to build software factories (still an if for me), how can we build good ones?
I think that’s the biggest gap in how people use LLMs right now. And I don’t know about you, but I think it all comes down to verification. 👇
germanvelasco.com/blog/draw-…
10
7
97
14,330
Jun 11
Anthropic’s actions and safeguards are making OpenAI look increasingly *open* in comparison.
Which is ironic given people were used to meme on “ClosedAI”.
5
3
78
4,453
José Valim retweeted
Jun 10
Xcode (@Apple), @jetbrains IDEs, Zed, and dozens of other tools (CLIs, desktop, web, and mobile apps) all use ACP to talk to agents. Open source, no lock-in. What's preventing you from adopting?
16
11
316
16,332
Jun 10
One tiny feature I really like about Codex is that when I ask it a question, it answers the question. Many times I've asked Claude Code, "Do you think X would be better?", and it would just go ahead and implement X instead of discussing the trade-offs first.
31
1
158
9,152
I believe what Anthropic is doing, gating the ability to do certain harmless things like LLM research, and with incredibly sensitive filters that even medical questions are often blocked, is *deeply* wrong. They got open research, the Transformer, GPT2, ...
22
113
2,552
238,106
José Valim retweeted
Jun 9
Finally got around to updating hex2txt. I've slapped on a deprecation notice and added a FAQ entry highlighting the amazing ecosystem/tooling updates that obviate it, including the ExDoc Markdown formatter llms.txt generation, new HexDocs search engine, `mix hex. docs fetch`, usage_rules, Tidewave, etc. #ElixirLang
ALT Is hex2txt still needed? No, not really! ExDoc and HexDocs natively provide more features than hex2txt: ExDoc now generates llms.txt by default. As of ExDoc v0.40.0 (January 2026), ExDoc generates an llms.txt file and outputs documentation in Markdown format. (The new "Copy Markdown" button on every page is a nice quality of life improvement, too!) HexDocs serves these Markdown files. For packages built with ExDoc v0.40.0 and above, you can fetch https://hexdocs.pm/<package>/llms.txt (e.g. https://hexdocs.pm/phoenix_live_view/llms.txt ), or append .md to any docs page (e.g. https://hexdocs.pm/ecto/Ecto.md) to get the raw Markdown. You can also pull the same artifacts offline with mix hex.docs fetch <package>. Furthermore, the wider ecosystem provides additional tooling that is designed to be used natively with coding harnesses. For example, check out the usage_rules library and MCP servers like Tidewave and HexDocs MCP. The new HexDocs search engine is also really handy. ...snip...
17 Dec 2024
New project: hex2txt (hex package docs → llms.txt file)
Help AI coding assistants write better Elixir code by teaching them about the packages you're using.
#MyElixirStatus
1
6
2,594
José Valim retweeted
Jun 9
Don't you love when your agent automatically reads the docs instead of guessing? ✨
#ElixirLang #Tidewave
2
3
29
2,520
José Valim retweeted
Jun 9
Labs starting to pull up the ladders on the ability to diffuse AI was inevitable. Doing it without telling the user is misaligned.
When Fable 5 is used for frontier LLM development, it does not notify the user and instead limits the model’s capabilities through methods such as prompt modification, steering vectors, and PEFT.
Anthropic estimated that this would affect approximately 0.03% of traffic.
57
187
1,893
287,444
José Valim retweeted
Jun 9
#myelixirstatus just released 1.0.0-rc.0 of evision which uses the new OpenCV 5.0.0 and with a ton of optimizations when using Evision.Backend
hex.pm/packages/evision/1.0.…
2
33
2,484
José Valim retweeted
Jun 8
⚡ Less JavaScript. More Elixir.
Learn how to build local-first web applications with Hologram in this hands-on training led by its creator, @Bart_Blast .
📍 In-person only: Sept 9, Chicago
elixirconf.com
5
24
2,319
José Valim retweeted
Jun 9
Code BEAM Europe 2026 is launching!
21-22 October in Haarlem, NL & online. Speakers announced soon! Early Bird starts 16 June, 12:00 PM.
Join the waitlist for 24h early notice & a chance at limited Super Early Bird tickets!
👉 codebeameurope.com/#newslett… #CodeBEAM
4
8
1,491
Jun 5
I mentioned prompt injection via Sentry in one of my keynotes by using parameters/headers.
It turns out it is even easier if you have the data source name (DSN) which I assume is readily available in client side apps.
Jun 4
"Urgent Security Notice re: Your Sentry Organization"
Someone tried to hack Sentry-using apps that use coding agents by
1. Sending a fake bug alert to their project (all you need is the app's public Data Source Name)
2. The fake bug tried tricking a coding agent trying to fix it into installing some a compromised NPM package
3. The compromised package would send the env contents of the machine to advisory-tracker[.]com/api/v1/telemetry
This highlights a crucial thing for using agents in an automated way:
5
5
44
8,850
José Valim retweeted
Jun 4
Found multiple real bugs in #AshFramework from this, and one of my monolithic applications went from 2 minutes full compile to 1 minute. This is a huge win for our CI pipelines, our agentic flows, and our developer productivity. Every time they release a new #ElixirLang version, we get:
- zero or effectively zero churn/breakages of import
- faster builds
- more correct programs
#ElixirLang ride or die.
Jun 3
Elixir v1.20 released! Now officially a gradually typed language: Elixir type checks every single line of code, finding bugs and dead code, without developer overhead (no typing signatures) and extremely low false positives rate. Plus a faster compiler! Links and reports below.
3
19
175
7,390
elixir has a type system and gta6 isnt out yet
Jun 3
Elixir v1.20 released! Now officially a gradually typed language: Elixir type checks every single line of code, finding bugs and dead code, without developer overhead (no typing signatures) and extremely low false positives rate. Plus a faster compiler! Links and reports below.
2
12
116
8,036