diffie.ai QA E2E testing agent
In SF CA, from Coimbatore
- Tweets 3,469
- Following 828
- Followers 883
- Likes 5,552
damn, 100k views on this bad boy
browserbase is dead
you can spawn 2k concurrent 24/7 VMs on boat.dev and spin up to 29M browser sessions 3 min long each, per month, for just 26k
vs 155k on them
used it to scrap 366 products & their funding for battleships.dev
3
1
26
1,862
is it cheaper to spawn a boat.dev vm and use a browser? than just use browserbase directly?
1
1
21
recommended reading, though i still disagree, as explained on @syntaxfm early this year.
for code, both memory and docs are not efficient. your codebase is all you need. make it modular, make modules small, at most, have a little maintained map of where what is.
liao.gg/blog/agents-dont-nee…
32
23
405
18,432
It’s happening guys. The saaspocalypse.
The ability to own context _even if_ your personal clone is inferior is better than amaze SaaS tool but silo-ed context.
Salesforce charges us $18,000 a month. So I built a replacement.
It started as a personal project: one day in Claude Code on the UI. The code quality was poor, so Orchestrator cleaned it up for four days. $5,600 in tokens. It's live.
Guess which one we're keeping.
2
4
1,891
Exactly. But to be precise, I meant more context that you control vs context that is not in your control. Less local vs API - that's an independent thing.
1
48
Bangalore is actually insane man
met a founder (@ayushjha__) with just 3 employees
all 3 of them live with him in his 2 bedroom apartment
they work 6-11-7 6am to 11pm, 7 days a week
they need his permission to do basically anything that isn’t work or sleep
he didn’t want them wasting time dating outside the company either, so he started matching employees with each other
80% of the team is guys, so apparently a few of them are now gay
Bangalore startup culture is on another level man
106
15
822
87,235
lol.. the first half felt a bit like how @hackerrank started. while it was interviewstreet. in adugodi bangalore . 15 years back.. @sp2hari @suren
2
6
848
Dang again !!! fable 5.1 said nothing was running. i showed it the evidence. then: "you're right, and i was wrong."
favorite model. still have to check its work everytime now!
74
yes in 2011, for IPC
web.archive.org/web/20110417…
haha.. @sp2hari
3
193
He needs to teach me how to use it better. For now it hallucinates does useless stuff too much. Using astra on high mode. And so damn slow. Runs 500 tests it doesn’t need. Claude just get it.
2
134
I've actually been using Pi Durable in Halo with a custom storage adapter for @tursodatabase
It absolutely tears
People of Pi: We've shipped Pi 1.0 with Pi Durable. Go make them yours.
earendil.com/posts/pi-1-0/
2
5
547
Introducing Pi 1.0 for Rivet
Running agents inside the sandbox is simple in theory
But in production, the sandbox's blast radius becomes your agent's blast radius:
💥 Sandbox OOMs → agent dies
🔑 Credentials → need a proxy
💤 Sandbox asleep → no crons
🔄 Update → redeploy every VM
Now with Rivet, you can run the vanilla Pi harness in your backend, separated from the sandbox
The sandbox can crash, sleep, or fill its disk. Your agent keeps running
Built for background & multiplayer agents
6
14
172
20,934
Vanilla Pi, no wrapper
We believe harnesses are opinionated and that matters. We expose vanilla Pi directly to you and handle the orchestration.
- Session methods are actions: 𝚙𝚛𝚘𝚖𝚙𝚝, 𝚜𝚝𝚎𝚎𝚛, 𝚊𝚋𝚘𝚛𝚝
- Pi events stream to clients unchanged
- Custom tools via Pi's 𝚍𝚎𝚏𝚒𝚗𝚎𝚃𝚘𝚘𝚕
- Session tree saved to SQLite
When Pi ships, you get it.
2
5
459
@NathanFlurry I am building a cloud agent with an architecture inspired by your harness outside the sandbox article.
However I chose to store all the llm session data in a central db, and not a db per actor.
I am curious why rivet actors each have their own db. your thoughts?
11
Do agents really need a CLI now ?
With ultrafast browser use with Astra or Jev, my Codex is browsing the web and clicking buttons.
CLI is fast, easier to test, easier to write evals against, deterministic. Yes.
But LLMs are exceptionally inteligent.
1
1
81
From founder point of view, venture capital has never been as risky a bet as it is today.
The entire funnel to an exit is narrower and longer. The bar to raise at each stage is higher. The number of companies that can raise at every stage is lower. Timelines to exit are longer. Non-consensus bet-takers are scarce.
The amount of money you can raise is more, which helps, and the TAMs of available markets are bigger.
But, the growth curve you as the founder have to create is much much steeper.
Venture capital's existential issue:
- One end of the venture market is small, focused on discovery, and relies on exits.
- The other end is large, focused on winners, and relies on generating markups at scale.
Growth of the latter has come at the expense of the former, by systematically delaying or destroying liquidity.
This imbalance is cannibalising venture capital's ability to interface with new ideas and talent.
It's why the largest firms are all concentrated in the same themes (often the same companies), and are even recycling the same founders.
The narrowing market only increases the velocity of capital, masking the problem. New seed rounds are falling, but Instinct can still raise a $1B on $10B.
The solution is to look at vehicles that can allocate capital from large LPs to both ends of the market, fuelling the growth of small and emerging managers.
This points to data on the marginal outperformance of fund of funds, and evidence that (unlike venture capital) they offer a truly scalable product — providing a better solution for large LPs.
Read more: credistick.com/the-illiquidi…
1
1
4
797
Linear and Notion come to mind. If I squint they are places a company’s work can be tracked (okr-ish) … both billion+ and technically clone-able with agents. Yet, no serious challenger in the last year. I also don’t know of any major company that have replaced them with an internal vibe coded clone.
A new competitor to them will have to solve differentiation and distribution, but if they do I could see another billion+ clone
1
30