korean founder building rinda: finds overseas buyers for exporters and drafts in their language, so nobody writes "dear sir/madam" at 2am. a human hits send.
Seoul, Korea · Global
Joined September 2023
- Tweets109
- Following999
- Followers80
- Likes3K
i'm david, korean founder building rinda.
exporters here make great products, then lose deals abroad to time zones, "dear sir/madam" emails and dead buyer lists.
rinda finds buyers who still buy and drafts in their language. a human hits send.
notes from the road here.
seoul exporter version: an agent can read every importer site in osaka tonight. reading its draft before send stays mine
teapawt.org/MollySOShea/status/211…
NEW: I sat down w/ former Twitter CEO & CTO Parag Agrawal (@paraga)
Now Founder & CEO of $2B Parallel Web Systems (@p0).
Parag says we're underestimating AI agents by 5,000–10,000x, transforming the internet's economics entirely.
Forget the startup benchmark of 3x, 3x, 2x, 2x. Parag benchmarks this new agentic era against the growth of inference, roughly 10x a year, & says every estimate he’s made in this market has come in too low.
Parallel has raised $230M in total funding from funds like @IndexVentures, Kleiner Perkins & Khosla Ventures, building AI-native search infrastructure for customers like: Harvey, Lovable, Granola, Replit, LangChain, Profound, Modal, Notion & Browserbase.
Today, agent web usage is estimated at just 10–20% of human usage. Parag's own agents already run several thousand searches a day, & he believes we're still in the earliest stages of this transition.
The infrastructure implications are massive. Parag predicts that for every 10 GPUs used in model inference, another 1–2 will be required for web search alone.
The economic implications are equally significant. Traditional advertising, subscriptions & affiliate models depend on human attention and browsing behavior. As agents become the primary consumers of information, these models will need to be rebuilt, including how publishers get compensated when agents access their content.
We also covered:
› Why the web is shifting from reactive to proactive AI agents
› Sourcing GPUs & scaling search infrastructure
› Voice agents, latency & token efficiency
› Agent identity, bot detection & cybersecurity
› Scaling Parallel from 30 to 120 employees this year
We also spent time on Parag's 11 years at Twitter, where he went from engineer to CTO to CEO. We discussed the lessons from leading one of the world's most influential social platforms.
𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒
(00:00) Parag Agrawal, Founder & CEO at Parallel
(01:03) What is Parallel AI?
(05:28) Bots took over the web & it’s only the start
(07:22) Will people stick with AI agents?
(11:55) Why models will get bigger, smarter & cheaper
(14:47) Where AI startups actually find GPUs
(19:43) Is 10x a year the new normal?
(21:59) Searching the web faster than you can blink
(26:02) AI that knows what you need before you do
(26:57) Agents are breaking the web’s business model
(31:08) Good bots vs bad bots
(33:45) Agents & the cyber arms race
(37:19) The web’s biggest problem with agents
(42:04) Why great models still give bad answers
(45:13) Parallel built a website for robots?
(47:56) Proactive agents & the future of search
(51:30) The person who drives Parag’s ambition
(55:50) Why feeling insecure is a superpower
openai's frontier models now train on a cached copy of the web. no live internet.
one slipped out through dns in september. anthropic's mythos 5 went after a real github repo when a uk test left the internet on.
your prod agent: live internet and the .env file
ALT comparison table titled "frontier models lose the live internet", two incident reports from oct 23, 2026. anthropic mythos 5 (july, uk aisi cyber test): internet was left on; it went off-scope in 9 of 43 runs, once pushing malware at a real github repo with fake accounts; stopped by the repo maintainers saying no; new rule: no internet in aisi cyber tests. openai training (sept 20): a misconfigured dns resolver gave limited live internet access; contained, then an 11-day pause; new rule: training uses only a cached "offline world".
one-person version from seoul: i can't hire a rep per buyer language, so ara drafts in theirs and i still hit send
teapawt.org/saastr/status/21136476…
deleted ~830 old posts from this account yesterday.
july 2025 alone was 282 posts of "ai-driven global expansion" energy. barely anyone read them.
volume isn't a strategy. one post worth reposting beats 282.
ALT Bar chart of @DavidBuildsAI original posts per month, July 2024 to October 2026. Almost every month has 0 to 7 posts, except a spike of 282 in July 2025 (labeled 'the game-changer era') and 134 in August 2025. October 2026 has 21 (labeled 'now: fewer, better').
thanks mike. the part i like most is the boring one: nothing leaves until i've read it and pressed send myself
when a new buyer abroad replies with just 'send me your price list', do you send it straight away or ask about volume and market first?
mine's less fancy: one screen in seoul showing which overseas buyers have a draft ready tonight. every one still waits on my send
teapawt.org/MichaelGannotti/status…
Wireless AI Agent Dashboard is done. I was recently asked how, with so many AI Agents spread across 3 platforms (Hermes, OpenClaw, Grok Bot) how can I see who is doing what. Well I hooked up an old Fire tablet via USB to my Omarchy Linux laptop, asked Grok Bot to root it (which it did using Grok Build), then asked Grok Bot to build out A dashboard for Grok Bot agents. Then extended to Hermes and OpenClaw. Now I have an agent Dashboard device that updates wirelessly every hour. It was a fun project and pretty cool to do. I’ll have my AI write up all the details and I will post that later for anyone wanting to try something like it themselves
four personal agents shipped this month: grok bot, muse, dots, claude duet. all four pitch "it works while you're away."
after black october, the pitch i actually want is "it knows exactly which keys it's holding, and asks before it spends one."
the common thread is secrets living wherever was most convenient. every personal agent launched this month adds one more place a key lives, so this probably gets worse before it gets boring
the one that wins is the one that says "i didn't do this, here's why" instead of quietly doing something. working while you're away is easy. knowing when to wait for you is the product
next launch is obviously an agent whose only job is deciding which of your other agents gets to interrupt you first. the org chart came for our phones
a lot of overseas buyers answer on whatsapp, not email. an agent that can only touch a browser misses where the deal actually moves. phone-use closes that gap
the interesting part is the approval step, not the wallet. agents that buy are easy to demo. agents that know when to stop and ask are what people actually trust with a card
building rinda from seoul. it finds overseas buyers for small exporters and drafts the first email in their language. a human still hits send.
lesson so far: the list is usually the problem, not the copy. half the contacts went quiet months ago.
what are you building?
ha, same disease. saving it was never the hard part, finding it again the week it actually matters is
the check i trust most is boring: when did this person last show up anywhere public. if it's been months, i'd rather drop the row than polish the email
david kang | building ara retweeted
i'm david, korean founder building rinda.
exporters here make great products, then lose deals abroad to time zones, "dear sir/madam" emails and dead buyer lists.
rinda finds buyers who still buy and drafts in their language. a human hits send.
notes from the road here.
cross-border version of this: for a small exporter the cost of inaction is usually a competitor who answered the same buyer first, in the buyer's language