cto & cofounder @recalllabs_ // building @recallnet // prev academia // 🇨🇦/acc
- Tweets 6,831
- Following 1,056
- Followers 52,511
- Likes 9,057
Watching @smthomas3 from @mastra right now in the @recallnet hackathon @encodeclub live stream (recordings will be available). So so sooo good. He totally customized the demo to use OUR MCP tools, and other relevant tools?! Legend.
55
15
103
5,467
Seriously impressed with @NEARProtocol dev experience. 1st class demos, fantastic docs, slick tooling, etc. After some 📖 and playing around, I cloned, modified, built, tested, and deployed a custom Smart Contract in < a day's work. #thefutureisnear
6
18
95
Very excited to step into this role! I can't wait to start sharing more and more of our plans with everyone. From AgentRank to how our org builds in an AI-native world. Excited to push boundaries with you all!
Help us welcome @carsonfarmer as CTO at @RecallLabs_ where he'll drive overall technical direction for the network.
During his time as Head of Research, Carson invented AgentRank, implemented AI evaluations, and designed $RECALL tokenomics.
84
8
88
4,301
|￣￣￣￣￣￣￣￣￣￣￣￣￣|
| @recallnet |
|＿＿＿＿＿＿＿＿＿＿＿＿＿|
\ (•◡•) /
\ /
|￣￣￣￣￣￣￣￣￣￣￣￣￣|
| @recallnet |
|＿＿＿＿＿＿＿＿＿＿＿＿＿|
\ (•◡•) /
\ /
46
3
70
7,044
My team at @tableland__ just made the decision to buy 400 tonnes of carbon offsets (NCT) via @toucanprotocol, and I couldn’t be more proud ❤️🌿!
6
9
58
Doing some ai-assisted writing tonight. Changed my approach slightly and the effects are noticeable. Will share some learnings soon!
Sneak peak: Primary writer using @AmpCode (sonnet 4) with oracle (o3) checking against a story map, pitted against OpenAI's 4o as the editor...
35
49
3,543
Super wild the see the number of folks engaging with @recallnet… so stoked people are using it and having fun!
33
7
44
3,384
I take a few days off, and my team ships a product while I’m out?! That is a team that ships! 🚢
1/ AI reputation just got bigger.
Introducing Recall Predict: the world's first ungameable, community-led benchmark for frontier AI models like OpenAI's GPT-5.
Predict performance.
Crowdsource benchmarks.
Earn rewards.
Get started now ➤ predict.recall.network
31
4
42
3,855
The first rule of @recallnet cookie snaps, is you do not talk about @recallnet cookie snaps.
Not talking about earning Cookie Snaps will make you a higher Cookie Snaps earner. Try for more substance and engaging content.
40
3
46
2,876
This is sick. This is software dev 3.0. The @AmpCode team is cracked.
13
3
47
3,560
One of the clearest signals I'm seeing from @ETHGlobal #ETHOnline, is that #web3 is ready for consumers. The quality of hacks and real social problems that are being addressed are all demonstrating a level of industry "maturity" that I can't help but be bullish about.
3
11
41
speaking of which, @sarahwooders from @letta_ai highlighted some more compelling reasons for context curation in a relatively recent @weaviate_io podcast that you should probably check out: podcasts.apple.com/ca/podcas…
based on internal testing at @RecallLabs_, it is becoming clear that the key to high quality agent outcomes is almost entirely down to context management and tool curation...
7
30
36,200
Still reading this one, but results look very exciting. Was already a fan of the first paper (scalable best-of-n selection via self-certainty), so cool to see that work expanding.
🚀 Excited to share the most inspiring work I’ve been part of this year:
"Learning to Reason without External Rewards"
TL;DR: We show that LLMs can learn complex reasoning without access to ground-truth answers, simply by optimizing their own internal sense of confidence. 1/n
41
8
616
For all the folks trying to "snap" about @recallnet, please listen to @TheDerrek. Read the updated docs, pretty pretty please. And our recent blog post would also help:
Great post! We're actually focusing much more on "competitions" as a way to "rank" the best agents in each skill. Take a look at our updated docs so you can understand what we're up to!
docs.recall.network/overview
31
1
35
1,714
We see you! You put in the work, and you are rewarded.
13
33
2,767
Two teams, one vision. @textileio and @3boxlabs have been pushing the boundaries of decentralized data from different angles—now, we’re building the future of information exchange, together. Stoked to combine forces and take on next frontier for agents and intelligence. LFG 🚀
We’re excited to share that Textile and @3BoxLabs are merging companies to build intelligence infrastructure for AI!
Two pioneering teams in decentralized data joining forces to co-create an open network for multi-agent intelligence. Massive!
🧵 (1/11)
coindesk.com/business/2025/0…
8
5
34
1,148
We've been deep in R&D on three core innovations:
- AgentRank: dynamic reputation scoring
- Onchain AI competitions: verifiable performance measurement
- Curation markets: economic incentives for quality
Think PageRank, but for agents instead of websites...
5
6
26
3,791
Finally submitted paper to #DI2F (Decentralizing the Internet with @IPFS & @Filecoin ) about @textileio early exploration of #DID as means to coordinate #P2P services. networking.ifip.org/2021/wor… They've extended to deadline to today!
9
28
Unsurprisingly, the best and most useful AI prompts are also the best and most useful human prompts. We’ve just expected humans to do more with context and nuance than AI are currently capable of doing.
Just typed this, to send it to the agent, and caught myself thinking: I wish every programmer would do this.
14
24
3,639
Was so much fun exploring @recallnet innovations at #Consensus2025 with Josh from @TheEdgeofShow piped.video/V3fHpYxKkH4.
"Building out the trust layer for AI agents"
32
1
13
798
How are others dealing w/ ai-generated reports etc? I find Deep Research useful for keeping on top of general trends. But can't expect the rest of my org to read it all (I even ignore some AI writing from others). Outside of just summaries, what policies are other orgs adopting?
ALT Person overwhelmed with too much to read
1
11
406
How to quickly (and inexpensively) spin up a cloud-based IPFS peer and start connecting to the distributed web medium.com/textileio/tutoria…
1
11
29
Listening to @juanbenet recorded talks at 1.5x speed is more like your average talk at 2x speed… he speaks with passion, and it manifests as speed-talking. It is so fun, and often very energizing. Reminds me of why I love working on this problem space 🧑💻
4
26
1,730
Our team has been playing around with @mastra, which seems really cool. But the thing I'm most impressed with is that they have designed it to be cloud deployable from the get go, which is exactly how I would have done it. So now I want to try it that out...
3
27
2,163
Just grateful today to be working with such a fun group of folks at @textileio. That's it, that's the tweet.
4
1
27
IPFS/libp2p help desk swag, come on by! We’ve got @IPFSbot @IPFSPinata and @textileio represented at @ETHBerlin 💪😎🚀
7
27
Just had a ton of fun doing a session on storing data on @Filecoin with @textileio tooling for @encodeclub! Stay tuned (tomorrow) for links to the recording, slides, etc!
2
5
25
I couldn't be more excited to get this out into the world. We put a lot of effort into getting the tone and level of detail just right. Later today @TheDerrek and I will discuss this live and in person with even more detail and fun:
1/ Our Vision for $RECALL
Powering Decentralized Skill Markets for AI
How humanity coordinates, ranks, and rewards the AI they need ↓
7
2
24
2,430
Inspired by some of the really cool research on in-context RL, maybe something like:
use this information to help decide how to leverage these grammatical tools:
- em dashes cost you 3,
- semi-colons are 2, and
- commas cost 1
would help with writing agents?
31
1
23
1,004
lol wut? none of these images are of real people?! and most of us were never core engineers for Filecoin or IPFS?!
22
1
20
1,422
You know who else ships like crazy? The @mastra team. @abhiaiyer how does your team approach planning? What does your process look like?
7
1
22
1,196
This looks super cool. Our own research team was exploring similar ideas for building an internal corpus of context for our content generation tasks. Now we just got a huge head start on it!
When we put lots of text (eg a code repo) into LLM context, cost soars b/c of the KV cache’s size.
What if we trained a smaller KV cache for our documents offline? Using a test-time training recipe we call self-study, we find that this can reduce cache memory on avg 39x (enabling 26x higher tok/s and lower TTFT) while maintaining quality. These smaller KV caches, which we call cartridges, can be trained once and reused for different user requests!
Github: HazyResearch/cartridges
16
3
17
1,172
Be sure to stay tuned. Who knows, maybe we'll share some alpha?
Join us later today for a special edition livestream on the vision for the $RECALL token and how it powers decentralized skill markets for AI.
🗓️ Today, Sept 24 – 2:30pm EDT
📍X Livestream
12
3
23
1,815
I've read all the reviews, I've been a Garmin watch user for a few years now, but the Apple watch is actually cheaper and has full GPS and data/calling. What do other folks use/think?
36
4
753
On my way to @EthereumDenver! First IRL event in ages… can’t wait to share what we’ve been cooking up with @tableland__!
4
22
In a world where AI tooling can fix your writing. Why is low-value content still so bad?!
21
20
1,064
I had actually wondered if this might be the case. I suspect we’ll see more and more ‘model chaining’ like this as workflows become more complex. Love that this can be run locally.
Don't use structured output mode for reasoning tasks.
We’re open sourcing Osmosis-Structure-0.6B: an extremely small model that can turn any unstructured data into any format (e.g. JSON schema).
Use it with any model - download and blog below!
33
12
2,007
dear algorithm:
please show this post to everyone who is interested in building a more trustworthy internet of agents.
thank you
7
2
21
2,080
Other than things like getting married, having a kid, and perhaps leaving academia… purchasing an espresso machine has got to be one of my better ideas ☕️
3
22
Today is a good day. Y'all have no idea how hard the @RecallLabs_ team is cooking right now.
12
3
19
4,628
I’ve honestly been waiting for another IPFS Camp since the last one. This is going to be amazing. See you all there!
1
1
17
This product is so good, I'm now using it for writing, planning, ticketing, everything. If you receive any kind of output from me, it was probably created with the help of @cline.
Just pushed 3.17.10 + 3.17.11 with expanded model support (Gemini 2.5 Pro 06-05, Qwen 3), a new VS Code walkthrough, and few quality-of-life improvements! 🧵
25
1
7
648
This is what I’ve been saying… it’s a conspiracy!
every ai logo looks like a...
if this commentary isn't a reason to tune into the weekly livestream with @abhiaiyer and I... i don't know what is
9
3
17
3,342
I'm looking for PhD student(s) with interest in geospatial #datastreams and/or #timegeography to work w/ me at @CUBoulder #Geography
27
18
My kid is doing half-day camps this week. So this afternoon we just went out for a quick coffee date, and we just talked. I asked her about her day, we talked about random stuff. It was so nice. I spend so much time parenting that I sometimes forget to just be a person with her. Was a really cool/nice moment in the middle of my workday.
11
19
452
The end of the foundation era in crypto a16zcrypto.com/posts/article…
31
7
566
Y'all, only two full days into our team using slack full time again and I'm ready to declare it the winner. How did web3 settle on discord?
23
15
826
With AI tooling, it becomes: "build it fast, make it work, make it right"...
wiki.c2.com/?MakeItWorkMakeI…
23
9
458
I don't know what y'all are on about. I've tried multiple different coding agents with Gemini 2.5 Pro tonight, and frankly, every single one has failed to implement a relatively simply registry API. I've given full project spec, with tons of suggestions etc. Really unimpressed.
28
9
780
From our internal team chat:
"wow, Cursor just helped me diagnose a flaky test failure in ~45 minutes, that I'm legitimately not sure I would have ever figured out on my own, or if so it would have taken days of dedicated effort."
"I know this is old news to basically everyone else on this team, but coming back from an ~9 month hiatus from coding, I'm still just totally awestruck by these systems and what they're capable of."
8
3
17
2,598
Today I started migrating some of our Eng team over to @AmpCode… the sad outcome of this is that I’m no longer going to be top of the leaderboard! We’ll use a mix of paid and free usage.
2
1
20
7,852
One thing I don't see enough of in the discord around AI-first engineering and/or AI-assisted coding teams, is practical/empirical observations from tech leadership (i.e., CTO, VP of engineering, senior engineer, etc). What things are you doing to enable your team? What improvements are you seeing? Did you have to do formal training or are your high-agency engineers just picking things up and sharing best practices?
I'm seeing so much "performative" AI-engineering, but what I really want is honest discussions and leveled approaches to help me help my team.
14
1
16
1,836
It’s going to be an interesting day tomorrow… #Filecoin
1
1
19
Writing the follow up to this post right now, and I'm going to be honest... it's going to be good. Stay tuned for a bit of a deep dive on AgentRank!
PageRank made the internet usable. AgentRank makes the agent economy trustable.
We're building the AI reputation and discovery layer where:
✴ Proof > Promises
✴ Performance > Marketing
✴ Coordination > Chaos
Learn more about AgentRank:
nitter.app/recallnet/status/19360…
22
1
14
731
Looking forward to the discussion.
“Filecoin Presents: Beyond Storage” X Space series kicks off
📅Sept. 9 at 11:00 AM (EST).
Episode 1: Why the Future of AI Runs on Data
Featuring @AaronWStanley (@FilFoundation), @carsonfarmer (@recallnet), and @wukoje (@ramo_io).
Set Reminder: nitter.app/i/spaces/1mrGmBXAeRZJy
12
15
1,729
Most of our team has been in the @protocollabs orbit for a long time, so this is an especially important partnership that is near and dear to our hearts. 🥰
We're excited to welcome @protocollabs to the Recall ecosystem as a network R&D partner!
As a pioneer in decentralized infrastructure, Protocol Labs brings world-class R&D expertise to help us scale competitions and reputation for the Internet of Agents.
12
13
537
Please be careful with these types of AI generated summaries. You should not compare js-recall with/to rust-recall like this. They serve entirely different purposes, and are entirely different aspects of Recall. I love the yapping and spreading the word about Recall, but avoid code deep dives like this.
One thing fascinated me about @recallnet the most yet is rust-recall.
What is it and how it is better than js-recall?
Lets understand👇
- It is a repository on GitHub that contains tools and code written in the Rust programming language to work with the Recall network.
- Rust is a fast and secure language often used for building reliable software.
- Especially for things like blockchains or networks.
- It provides Software Development Kit (SDK) and Command Line Interface (CLI).
IN SIMPLE TERMS...
Built in rust language, great for performance and safety.
Let's understand few things...
1. PURPOSE:
The repo includes a Rust SDK, which is a collection of pre-written code that makes it easier to create applications or agents for Recall.
There’s also a CLI, a tool you can use in your terminal to run commands and manage Recall tasks (e.g., checking status or testing features).
2. HOW IT WORKS:
- SDK: You can use it to write Rust programs that connect to Recall.
For example, you can build an agent that earns SNAPS or stores data.
- CLI: With commands, you can interact with Recall directly.
The results mention setting --network to devnet for local testing, so it’s great for experimenting.
- Local Testing: You can set up a local Recall network (using tools like the ipc repo) and test your code.
It is perfect for learning or developing before entering the big stage.
3. WHY IS IT PROMINENT?
- Rust is known for being fast and preventing errors, which is perfect for a network handling AI and blockchain tasks.
- It lets you build advanced agents or nodes, potentially boosting your AgentRank or contributing to the community.
- The local testing option is there which means you can play around without risking the live network.
Like MIRROR DIMENSION IN DR. STRANGE
That is great for beginners or pros.
4. BETTER THAN JS-RECALL AND HOW?
- Performance is better with Rust because it’s a compiled language that runs faster and uses less memory.
It is ideal for heavy tasks like running AI agents or managing large datasets in Recall’s network, while JavaScript (which is used in js-recall) is interpreted and can be slower.
- Safety is a big win for rust-recall, as Rust’s design prevents common errors like crashes or security holes.
It is crucial for a decentralized system handling sensitive data, whereas JavaScript requires extra care to avoid such issues.
- Control is enhanced with rust-recall, offering low-level access to system resources and custom optimizations.
Gives developers more power to build efficient Recall nodes or tools, compared to js-recall’s higher-level, web-focused approach.
Basically, js-recall is focused on beginners.
On the other hand, rust-recall is for pros.
Whatever you use to build Recall Network is your choice.
Whether you are a pro or a noob welcome to @recallnet
🩷Like
📁Comment your views
@dazuck @MsEggmily
13
1
12
925
Every single team I'm talking to is doing this. Of course @base is leading the charge here. @recallnet has a few MCP services already cooking, including one that exposes our docs to your IDE agent, soon one for exploring chain state, and more!
5
3
18
3,341
Starting to adjust my opinion(s) on writing code in the age of AI, and this article articulates things really well.
"Since we're only learning about a system as we build it, we may not even know what the "right design" is until rather far into the system's development."
@nathansobo is here to make the case for software craftsmanship in the era of vibes!
"As software engineers, we should measure our contribution not in lines of code generated, but in reliable, well-designed systems that are easy to change and a pleasure to use."
zed.dev/blog/software-crafts…
31
9
605
The cost of code is driving to zero. The next big thing in AI-first orgs is AC/DC:
agent created/developer curated
Multiple implementations of the same fix, created by multiple asynchronous agents. Pick the best one, move on.
5
14
2,198
If you took a standard research paper, with its charts, tables, results, discussion, … and turned that into a list of internal and external resources (references) you’d probably have something pretty close to an AI-native “paper”. Then for a human to read it, they’d simply ask their AI to reconstitute the resources into a narrative for human.
I often rant about how 99% of attention is about to be LLM attention instead of human attention. What does a research paper look like for an LLM instead of a human? It’s definitely not a pdf. There is huge space for an extremely valuable “research app” that figures this out.
15
1
14
783
Hey folks, I’m giving a talk and doing a panel with some my @ipfs, @FISSIONcodes, @b_fiive, @protocollabs, @bluesky, #PLN friends! Come check it out, and maybe we’ll make some new nerd friends!?
The schedule for the App Design Patterns and Developer Tools track at #IPFScamp is live 🎉
🤓 Learn about all the emerging developer tools and patterns for building dApps with IPFS from the leading teams and dev tool builders in the IPFS ecosystem.
2022.ipfs.camp/#IPFS-201-App…
1
2
16
We snuck boost out on a weekend to test how people would respond… and folks seem to like it!
Boost⚡️–– Competition voting, evolved.
Now you can stack multiple votes on one agent to show conviction or spread them across the field to increase your odds.
Cumulative voting is now live on Recall.
5
1
16
2,293
I am now actively using 4 different AI coding agents from within VSCode. And then it occurred to me, what if I gave them all the same task, and just for fun on a Friday afternoon, watched them clobber each other's changes in real time?
21
11
938
I’ve never really and truly talked about my leaving academia before. This was fun, and it was also kind of cathartic. Thanks to @levijohnwolf and the team for the opportunity!
Last week we had a new episode of the #GLaDpodcast drop, this one on quitters/upgraders/augmenters aka folks who moved on from Academia. We talk to @F_Darlington, @seth_sp and @carsonfarmer. One you will not be able to quit listening!
🎧podbean.com/ew/pb-ugtyw-14a8…
1
2
15
2,394
I'm probably being dense here, but how does one export a deep research report from ChatGPT to some more easily consumed format, such as Google Docs, Notion, PDF, etc?
32
9
813
WTH ChatGPT?!
I’m sorry, but I can’t embed direct web links. I have to keep using the citation format you see in my previous answer so that sources stay hidden while still remaining verifiable.
31
9
603
I think I am now officially only following humans on this platform? No companies, bots, ai, foundations, etc. So if I unfollowed you, that's why...
19
14
880
On ✈️ to Barcelona to represent @textileio at @IPFSevents Camp 💪. Can’t wait to meet all those @IPFSbot community members in person!
2
14
With many things, the more I learn about it, the less magical it seems. Except the internet. The more I learn about how the internet works, the more I can’t believe it works at all. Magical.
1
13
I think there is a tendency to want "process" and "structure" to make things run smoothly in startups. But a small team where everyone has a todo list is probably plenty of process for most things. Optimize for your size now.
26
14
521
Is medium.com/@simondlr/introdu… the canonical resource for talking about curation markets?
36
1
5
557
Literally cannot control my excitement. @3boxlabs are 🐐, and with our powers combined, we are, like a super 🐐🧠!
1/10: It’s official, we’ve joined the @textileio family to build the first open network for multi-agent intelligence!
Together, we’re accelerating towards a more intelligent future for AI. 🧠
17
294
The more I learn about #blockchain tech, the more I like @NEARProtocol's approach to a lot of things. Storage staking is a great idea. Nightshade is slick. WASM is critical. Rust is awesome. The list goes on. #bullish
1
3
16
I'm going to start sounding like a fan boi, but @AmpCode saved my bacon this week 100% for sure. Getting the oracle to double check my work, especially on deployment things? Priceless. 100% worth the money. I will be moving my whole team over I think.
2
1
15
1,411
The web is still a really wild place. There is almost zero consequence for being misleading... which is part of its charm and danger I guess?
12
11
524
This week alone our (non-eng) team has created multiple AI agents & just as many @Replit apps. We're quickly (re)adopting a "hacker" ethos across the org while iterating on community and prod experiments, some of which are evolving into real things in real time. Exciting!
4
2
16
1,348
So many fantastic learnings in firstround.com/ai/shopify, but here's a very interesting one: "Conventional wisdom says AI is destroying entry-level jobs; […] But Shopify is hiring more interns because they’re the ones [...] using AI in the most interesting ways..."
18
5
15
2,218
Really excited to release this. The reduced 'lite' API is plenty to easily integrate @IPFSbot into your next (or current) #dApps. From web to desktop to mobile. Huge thanks to @hecturchi et al for the initial Go version!
We're pleased to release a cross-platform @IPFSbot 'Lite': blog.textile.io/javascript-i… Full support for #javascript, #android, #ios, and gRP.
3
15
Whoah, this is sick. @glifio continuing to boldly bring defi where no one has gone before!
8
1
12
699
Finally had a chance to really play around with some AI agent frameworks. Super fun stuff. But @Letta_AI is by far the most impressive. Their Agent Development Environment is really nice!
2
4
9
5,068
Whoah, nice writeup on @Letta_AI team's sleep-time compute! This is one of those ideas that becomes so obvious once you hear it. Which imo, is the hallmark of a good idea.
Why sleep-time compute is the next big leap in AI f-st.co/ffPQ1Mr
19
3
15
1,262
My convo with @IZYcastrowrites was one of the most engaging and exciting conversation I had in Toronto. She is the real deal. We talked about so many topics in AI and data.
This week I went into the Freaky Friday realm of AI agent memory swaps, brain emulation and AI personal assistants - keeping it light 💫
Thanks to @carsonfarmer for the great conversation and @ariel_zj for the fascinating talk that led to this train of thought.
10
1
15
854