Yi Ding -- prod/acc (@yi_ding) profile picture

Voice AI @valeriehealthHQ Prev AI Architect @qventus LITS and Partnerships @llama_index, Messaging Apps @Apple, HFT @ GETCO, @Citadel

San Jose, CA
Joined May 2008
Yi Ding -- prod/acc retweeted
I gave a fleet of agents 15 repos to run during my sabbatical. The slowest part of the whole factory turned out to be me. 📙 🚀 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝗶𝗻𝗴 𝗮 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗙𝗮𝗰𝘁𝗼𝗿𝘆, 𝗣𝗮𝗿𝘁 𝗜 is out today! 🚀 📙 The fleet included a design system, an auth service, a template repo, and a handful of customer apps and consulting projects. I measured what happened and wrote it down, including the parts that surprised me. There were a lot of learnings I had to share, and so I wrote a book! A software factory isn't a product. Like waterfall or agile, it's a way of building software. The book walks through its five flows: 1. Direction: deciding what the factory builds, from customer requests to production alerts to the agents themselves 2. Build: agents messaging each other, pulling in outside context, planning and writing the code 3. Delivery: agents reviewing the code against your instructions, then shipping it and checking that it works 4. Consistency: keeping 15 repos building things the same way, because a problem in one spreads to the others 5. Improvement: tracking cost, time and quality, and changing the factory based on what you see 𝗬𝗼𝘂 𝗮𝗿𝗲 𝗯𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝘁𝗵𝗲 𝗳𝗮𝗰𝘁𝗼𝗿𝘆, 𝗻𝗼𝘁 𝘁𝗵𝗲 𝗮𝗽𝗽𝘀. The book draws on my own battle scars, plus a lot of listening to AI Engineers solving the same problems. The agents may write the code, but people still decide what gets built. (Someone has to.) I've already started on Part II. If you're building a factory too, tell me what you've learned in the comments. I'd love to include your perspective. Link to the book in the first comment.
6
4
2
12
512
Yi Ding -- prod/acc retweeted
Replying to @yi_ding @benparr
small-to-big retrieval is underrated: embed the precise bit, then expand only after a hit
1
2
48
Yi Ding -- prod/acc retweeted
Unlike previous voice AI models, GPT-Live delivers both natural speech and strong intelligence. openai.com/index/introducing…
3
4
2
103
10,317
Yi Ding -- prod/acc retweeted
Maybe the one useful thing from the Omnigen release is it leaked Databricks's upcoming "lakebox" product (a competitor to @e2b @daytonaio is what it looks like). That would actually be useful to Databricks users.
1
1
422
Yi Ding -- prod/acc retweeted
Benchmarks place GPT 5.5 as the best model on SWE, but is it the best at making apps end-to-end? Turns out Opus 4.8 continues to be the king of vibe coding on both price & performance. Introducing ViBench: the first benchmark for app creation based on real world tasks
99
63
14
792
66,505
Yi Ding -- prod/acc retweeted
4,226 voices have already been recorded across 147 Humans in AI Week events happening around the world this week. I do not think I would have believed that three years ago. In early 2023, a small group of us were sitting in a cozy San Francisco apartment trying to make sense of ChatGPT, AGI, and what this all meant for the world. The technology felt historic. The thing I remember most was the feeling in the room. People were thinking out loud. Changing their minds. Admitting confusion. Getting excited. Getting scared. Finding language together. That night became @AICollectiveCo. Three years later, this community has grown to 250,000+ members, 200+ chapters, 1,700+ events, and 650+ volunteer organizers around the world. This week, that same instinct is becoming Humans in AI Week: June 1–7, 2026, across 100+ cities and 50+ countries. A global time capsule for the AI era. The question is simple: What does it mean to be human in the AI era? The answers are messy. That is the point. Some people are excited. Some are scared. Some feel behind. Some feel superpowered. Most of us are carrying a few contradictory feelings at once. That is why I still believe so much in rooms full of real people. Online, AI discourse collapses into extremes. In person, the temperature changes. People listen longer. The anxious parts get named with more care. The optimistic parts become less abstract. You remember there is a person behind every position. That has always been the magic of this community. We can gather frontier builders, curious newcomers, artists, students, founders, policy people, educators, and skeptics into the same conversation, then let the room do what the internet usually cannot: slow people down enough to hear each other. Humans in AI Week feels like the culmination of 3+ years of learning how to create those rooms. For me personally, this is also a handoff moment. I stepped back from leading The AI Collective because the community had become bigger than any one person, and because AJ, Catherine, and the team were ready to carry it into its next chapter. Watching them turn the original spark into something this global is SO special. Deeply grateful to @AJs_AI, @catrosemcmillan, our chapter leads, our volunteers, our partners, and every person showing up or adding their voice this week. This is what we built the community for. Onwards and upwards!! 🚀
2
8
12
708
Yi Ding -- prod/acc retweeted
I remember hearing @karpathy himself say that autonomous agents would, like self driving cars, take a decade to work at human-like levels. To see the Claude Code team deliver it in 3 years is both mind-blowing and just a testament to the continued exponential trajectory.
1
2
5
1,550
Yi Ding -- prod/acc retweeted
Claude Code is finally an RLM (oct 2025), congrats to Anthropic :-)
New in Claude Code (research preview): dynamic workflows. Claude writes an orchestration script on the fly, then spins up a large fleet of coordinated subagents in parallel to take on your most complex tasks. Use the word "workflow" in a prompt to get started.
18
42
6
491
80,285
Yi Ding -- prod/acc retweeted
Super excited to finally share Dynamic Workflows in Claude Code!! We built this a couple months ago, and it has slowly become a daily driver for a bunch of people at Anthropic. A few tips for getting the most out of it 🧵 teapaw.org/ClaudeDevs/status/2060…
New in Claude Code (research preview): dynamic workflows. Claude writes an orchestration script on the fly, then spins up a large fleet of coordinated subagents in parallel to take on your most complex tasks. Use the word "workflow" in a prompt to get started.
96
165
15
2,434
506,809
Yi Ding -- prod/acc retweeted
LiteParse is really neat! It does a great job of extracting text from annoying layouts in PDFs (multiple columns for example) It's only available as a Node.js CLI app, so I vibe-coded up this version that runs in a browser
LiteParse, our OSS document parser, is really good at parsing complex PDF layouts, text, and tables into a clean spatial grid. The best part is it doesn't use VLMs or any ML models at all. It's entirely heuristics based and super fast ⚡️ The secret lies in our sophisticated grid projection algorithm. This blog post by @LoganMarkewich gives a comprehensive walkthrough on how it works: 1️⃣ Sort lines based on similar Y coordinates 2️⃣ Extract left, right, and center anchors 3️⃣ Classify every text item into one of these anchors 4️⃣ Project every text item into a grid column (the exception is any paragraph of flowing text, which is rendered separately) 5️⃣ For any item projected into a grid column, that item is the forward anchor for all subsequent text items with the same anchor 6️⃣ Postprocess the final outputs to remove extraneous spaces and margins As an example, take a look at the results below. You can see text in the left column, with a nicely overlaid table on the right. LiteParse is fully free and open-source, you can use it today! Either directly through the CLI or integrated into your coding agent. Blog: llamaindex.ai/blog/how-litep… LiteParse repo: github.com/run-llama/litepar…
32
84
9
796
100,154
Yi Ding -- prod/acc retweeted
Sorry, this wasn’t intentional, text-embedding-3-small isn’t going away! teapaw.org/romainhuet/status/2047…
2
1
109
22,530
Yi Ding -- prod/acc retweeted
Today is a big day! We're launching a ~ new ~ version of Claude Code in the desktop app. It's been redesigned from the ground up for parallel work and is a lot faster. It's been my main way to use Claude Code for the last few weeks.
613
436
172
9,726
956,474
Yi Ding -- prod/acc retweeted
This matches my own experience with GPT-5.3-Codex – I basically only run it on xhigh nowadays for all coding tasks. With all the speed improvements too, it doesn't even feel that slow, even at `xhigh`.
decided to run 5.3 codex on xhigh as well, its 90%... rip IBench, survived 3 months.
13
13
5
277
273,068
Yi Ding -- prod/acc retweeted
For me, "Whatever @simonw posts" has progressively become my main way to track technical x product progress in AI (both open and closed, though i'd love to see a bit more open) His tech x product signal-to-noise ratio is *very high*
8
7
1
157
9,804
Yi Ding -- prod/acc retweeted
life is short build something that lasts
2
4
39
2,514
Yi Ding -- prod/acc retweeted
"The product sells itself" and "Sales is easier than building products" are my favourite laughable myths. PLG motion is excellent. Most companies should adopt it and do Growth initiatives. However: ▪️ Launching a product requires planning, good comms, excellent execution and a properly organised distribution. Posting on Linkedin & X with a nice video will not do anything. ▪️ If products sold themselves, Microsoft Hololens or Oracle Cloud would have been a hit already ▪️ Take Rabbit R1; it was an excellent campaign with lots of hype. Nothing is crazy good launching products, and Cursor & Lovable, the PLG kings. Salesforce is the best in B2B launch & hype. ▪️ Great sellers know what motivates their customers, how to reach their ICPs and when to be aggressive, be soft, go with the flow or just let it die. ▪️ Qualification is the toughest part of sales. You qualify every single time you outbound, negotiate, upsell, cross-sell or ask for feedback. ▪️ Complex industries are the toughest to crack; passion & drive gets you through the door but not a good deal.
5
3
29
2,954
Yi Ding -- prod/acc retweeted
Hence why decided to go all in on focusing on it
3
1
19
2,533
Yi Ding -- prod/acc retweeted
Do filesystem tools 🗃️ kill vector search❓ File search is all the rage right now. Give claude code access to the CLI and it turns into a generalized computer use agent. This is a great blog post from @itsclelia that benchmarks file search vs. vector search. Up until ~100 files, the speed is comparable. From 1k files onwards, vector search is way faster to get an initial result. File search does enable the agent to do dynamic reasoning to answer more complex questions (e.g. linking across different sections in the same doc). The real answer is probably blending both, ala what we did with semtools! github.com/run-llama/semtool… Check out our blog: llamaindex.ai/blog/did-files… We used LlamaParse to parse all files into LLM-readable formats. fs-explorer repo: github.com/run-llama/fs-expl… LlamaCloud: llamaindex.ai/signup
Can filesystem tools really replace vector search? We put agentic file exploration to the test against traditional RAG. Our experiment with fs-explorer agent vs. hybrid RAG revealed some surprising insights about when each approach shines: 🏃 RAG is faster - averaging 3.81 seconds quicker thanks to fewer LLM calls and consistent network requests 🎯 Filesystem agents are more accurate - scoring 2 points higher on correctness by accessing full file context instead of chunked fragments 📈 Scale changes everything - at 100-1000 documents, RAG outperforms filesystem exploration in speed and maintains quality ⚖️ Context matters most - filesystem tools excel with smaller files that fit in the LLM's context window, while RAG handles massive document collections The verdict? It depends on your use case. Filesystem agents work great for smaller, focused document sets where accuracy trumps speed. RAG remains king for large-scale applications requiring real-time responses. Read the full experimental analysis by @itsclelia and see the results for yourself: llamaindex.ai/blog/did-files…
13
20
3
185
26,017