
agent sandboxes will eventually be as ubiquitous and essential as the vscode was pre-2023, but probably less than 1% of developers have even experienced how much better working with it is. super exciting times ahead.
3
128
If you are technical and building developer tooling, first find pmf with yourself and if you find yourself using your own product with obsession daily for an extended period of time, that you'd genuinely pay for if it wasn't your own product, it is very likely that you are building something meaningful that a lot of people would use and pay for as well. However being honest with yourself is the hardest part.
5
260
Please don't spend $4k on a Mac Studio just for CI and local dev, its a beast of a machine but if you are running 10s of agents in parallel with hundreds of CI runs a week you will still reach full disk every few hours and it will still massively slow your computer down. May or may not be talking from personal experience.
6
347
Ive been tinkering with a new way of working with agents, b/c it sometimes felt like working with agents is a never ending cycle. if you are someone who puts in hours every day with agents and want to make sure your agents are not slacking or getting off track this will be for you. It has 10x my velocity and has brought 100x more structure while building remotehost.ai. more to come...
5
226
this is useful since the schema is persisted, but the agent session has real-time context of its active work, so you are not tied to a fixed set of args and you get best of both worlds.
use slash commands like they're CLIs, for example, I've got a `/table` slash command that works with my issue tracker backend that I created.
I can focus in on specific rows like `/table t1`, where t1 is row 1, which will render a scoped table of linear-like todos just for t1. Its got other commands like `/table active` to render all active rows currently in progress right now.
have been using this approach consistently as of recent and can't imagine not working this way.
3
232
use slash commands like they're CLIs, for example, I've got a `/table` slash command that works with my issue tracker backend that I created.
I can focus in on specific rows like `/table t1`, where t1 is row 1, which will render a scoped table of linear-like todos just for t1. Its got other commands like `/table active` to render all active rows currently in progress right now.
have been using this approach consistently as of recent and can't imagine not working this way.
CI as we know it is going to die.
Current approaches are too slow and expensive. The infrastructure and frameworks are fundamentally flawed for modern agent-driven development.
A lot of people here are trying to convince themselves that we don't need CI, because "agents" can do all the review, but more code output = exponentially higher risk and you will always need deterministic proof of correctness if you want to create reliable software.
Have been cooking up something for the past few months for remotehost.ai and I am really excited to share more about it very soon!
1
4
221
for productivity apps, I’ll never use an in-app chat feature. I will only use your app if it lets my agent use it.
1
3
168
sandbox startup time matters.
5 seconds instead of 50ms adds 4.95 seconds every time an agent starts a sandbox.
if your team starts 10,000 sandboxes a month, that’s 13 hours and 45 minutes of cumulative startup delay.
4
178
two of the biggest sources of compute spend in agentic development:
> the inner loop (e.g. agents repeatedly running builds and tests in parallel)
> ci
don't just "increase concurrency" like the sandbox companies here who are incentivized to tell you.
Start with these three things:
1. set up a shared cache across agent sessions and CI. If one agent already built or tested the same inputs with the same configuration, the others should be able to reuse that work.
2. make build and test targets fine-grain/hermetic. e.g. you can use something like bazel (created by google), which allows you to isolate affected work and deterministically cache the results instead of rebuilding and retesting everything.
3. increase concurrency of agent sandboxes + build/ci workers ONLY after you did #2.
4
7
482
I'm pretty sure traditional CI won't be around long. Even with heavy caching and other optimizations, some runs took hours. Have been tinkering with a new approach for a little bit. Going to give it more time on the clock before sharing it with the world, but so far it is very promising.
3
90
agentic commerce is coming along fast. had my grok bot use chat.opencommerce.sh to find me a mac studio on amazon in just a few seconds. congrats @keyserfaty and team on the launch!
today we are announcing openrouter for commerce: opencommerce.sh
a single endpoint that integrates 5k stores, including Doordash, Amazon and most shopify stores, with payments using Agentcard (Stripe Link and Shop coming soon).
we built opencommerce to solve 3 problems:
* coverage: one endpoint to integrate 5k stores. agentic purchasing is a core feature for most personal assistants, but not their main business. searching for products and buying on behalf of users shouldn’t be as painful as it is today (e.g. it shouldn’t involve browser automations or dealing with PSPs). opencommerce solves that: 1 tool, every online store.
* reliability: 100% uptime. the challenge with browser automations is that they fail (e.g. captchas, bot-detection, etc). we run thousands of tests every couple of hours on every one of the stores to guarantee that the integration works when your user tries to buy something.
* speed: responses in under 10s. consumers are used to Amazon speed, and agentic purchases have so far felt like a downgrade. most agents spend minutes finding products before they come back with results. we monitor speed and improve it constantly.
2
6
230






