I have still not found a better way to get good results from coding agents than to be hyper prescriptive, then keep an eye on what they're doing and interrupting them when they go down clearly wrong paths.
JLarky 3 0 0 19 1.7K 1
erik.eth 🛡
@programmer
Building something new.
Prev head of eng @CoinbaseDev, @ourzora
Creator of x402.org
15.9KFollowers
1.8KFollowing
3.1KPosts
299Media
14.7KLikes
We’re at the point where I automatically discount any product with a Claude default aesthetic landing page
5 0 0 42 2.9K 2
@programmer reposted
41 328 49 5.4K 513K 912
the cost per fixed amount of intelligence halving very 6 months is the Moore's law of our time
x.com/ValsAI/status/…
x.com/ValsAI/status/…
2 0 0 17 2K 9
I think we’re back to improvements to the harness mattering more than the model
1 0 0 18 1.7K 0
The gemini 4 results are honestly not surprising if true.
Gemini 3.7 flash was already a very good general purpose model, better than both fable and astra on tau 2 in both top end performance and price.
openrouter.ai/benchmarks/tau…
Gemini 3.7 flash was already a very good general purpose model, better than both fable and astra on tau 2 in both top end performance and price.
openrouter.ai/benchmarks/tau…
1 0 0 11 1.7K 1
Slop cannon or vibe snipper rifle. Choose your path young engineer.
8 0 1 23 3.3K 2
I'm still in stealth, but I'm hiring!
Location: SF preferred, but remote in North American time zones is possible for exceptional candidates.
Open roles
- Founding Engineer: build stuff
- Founding Product Marketing Manager: grow stuff
We offer generous salary, equity, health care, and the opportunity to work with smart and kind people on meaningful problems to help advance the state of agents.
How to apply:
Email jobs @ 5d.tech with:
- LinkedIn / X / GitHub / Resume / Personal site / etc
- A description of the thing you've built you're most proud of
- Your current favorite product & why you like it
- Your all time favorite website
- A good meme
Location: SF preferred, but remote in North American time zones is possible for exceptional candidates.
Open roles
- Founding Engineer: build stuff
- Founding Product Marketing Manager: grow stuff
We offer generous salary, equity, health care, and the opportunity to work with smart and kind people on meaningful problems to help advance the state of agents.
How to apply:
Email jobs @ 5d.tech with:
- LinkedIn / X / GitHub / Resume / Personal site / etc
- A description of the thing you've built you're most proud of
- Your current favorite product & why you like it
- Your all time favorite website
- A good meme
56 20 13 518 79.1K 336
"Have you tried {openclaw -> poke -> hermes -> grok bot -> instinct -> muse}?"
yes
yes
12 1 1 47 3.9K 1
impressive result on JevBench
Marco De Rossi@marco_derossi · Sep 23On JevBench, Sage1 scores better on accuracy than other System1 models, with a similar median latency.
When multi-step reasoning is required, Sage has 93% accuracy vs. 71% for other System1 models.
Sage is smart enough to know when to be fast and when to think step by step, covering both System1 and System2.
When Sage activates reasoning, it takes on average 3X longer to decide.
When multi-step reasoning is required, Sage has 93% accuracy vs. 71% for other System1 models.
Sage is smart enough to know when to be fast and when to think step by step, covering both System1 and System2.
When Sage activates reasoning, it takes on average 3X longer to decide.
Open quoted post →
0 0 0 6 2.2K 3
The internet is so underrated. Theres no chance we would have ended up with AI without the open web.
11 1 0 54 3.5K 1
@Must_be_Ash does it again
so sick!
so sick!
Ash@Must_be_Ash · Aug 25I pay tested thousands of x402 endpoints and I'm open sourcing the whole thing
think OpenRouter for tools meets n8n, served over plug-and-play MCPs
> you get a catalog of 2,000 services
> a visual workflow builder
> spend controls
> connect to Claude or ChatGPT with oAuth
think OpenRouter for tools meets n8n, served over plug-and-play MCPs
> you get a catalog of 2,000 services
> a visual workflow builder
> spend controls
> connect to Claude or ChatGPT with oAuth
3:57 8 3 0 53 5.9K 9
Agents are like water, they’ll flow to whatever solution is easiest
Erik Voorhees@ErikVoorhees · Aug 23Just had my first organic agentic x402 magic experience
• Needed my agent to access a website for which I didn't have an account.
• It discovered (on its own) Apify, which accessed that service on its behalf. Apify accepts x402 payments, which is what my agent then suggested
• Agent set up its own wallet (it has never done a crypto tx before), gave me a deposit address on Base. I sent it some USDC (and no ETH)
• Agent then connected to Apify on its own, arranged payment via x402, and got the data it needed, reporting it back to me.
The whole sequence involved just 4-5 single sentence prompts from me. That was all pretty magic, but then I asked it this question.
Agentic commerce on crypto rails is the future.
agent was pi agent
model was kimi-k3-fast-api on venice.ai/api
• Needed my agent to access a website for which I didn't have an account.
• It discovered (on its own) Apify, which accessed that service on its behalf. Apify accepts x402 payments, which is what my agent then suggested
• Agent set up its own wallet (it has never done a crypto tx before), gave me a deposit address on Base. I sent it some USDC (and no ETH)
• Agent then connected to Apify on its own, arranged payment via x402, and got the data it needed, reporting it back to me.
The whole sequence involved just 4-5 single sentence prompts from me. That was all pretty magic, but then I asked it this question.
Agentic commerce on crypto rails is the future.
agent was pi agent
model was kimi-k3-fast-api on venice.ai/api
Open quoted post →
15 6 0 107 10.3K 17
Of all the "manage many agents" tools i've tried @orca_build is definitely the best.
@zeddotdev for editing code / pair programming with an agent for high precision things, Orca for swarming.
github.com/stablyai/orca
@zeddotdev for editing code / pair programming with an agent for high precision things, Orca for swarming.
github.com/stablyai/orca
1 2 0 19 2.7K 5
Stacked diffs are now more useful than ever. Far easier to review LLM code if it’s broken into stacked diffs.
Gergely Orosz@GergelyOrosz · Aug 1Look, better late than never to ship stacked PRs.
But this feature was extremely useful 2015-2025 at most companies using it. Since ~Nov 2025 when AI models became more capable and write more and more of the code, stacked diffs have so much less relevance.
GH feels years behind
Open quoted post →But this feature was extremely useful 2015-2025 at most companies using it. Since ~Nov 2025 when AI models became more capable and write more and more of the code, stacked diffs have so much less relevance.
GH feels years behind
2 1 0 13 3.3K 1
fyi all @ChatGPT Codex tasks you start from @linear are default public & leak your internal repo + branch names.
Doesn't seem to be a way to force them private.
Doesn't seem to be a way to force them private.
3 1 0 17 3.6K 0
gotta say, grok build with 4.5 feels insanely fast
5 0 0 15 1.7K 0
biggest question of our time
1 1 0 17 3.6K 6
5 5 5 30 10K 8
every github repo being backed by a CD would actually explain a lot about github's performance & uptime
GitHub@github · Jul 2We heard you. And we agree.
In light of recent developments in physical media, GitHub is proud to announce that you can now obtain your public repo on CD-ROM.
Keep it. Lend it to friends. Pass it on to your children.
Your code is physically yours, forever. Until you lose it, let's be real.
Order yours today.
gh.io/cd
In light of recent developments in physical media, GitHub is proud to announce that you can now obtain your public repo on CD-ROM.
Keep it. Lend it to friends. Pass it on to your children.
Your code is physically yours, forever. Until you lose it, let's be real.
Order yours today.
gh.io/cd
0:14 3 0 0 16 5.1K 1