co-founder @ship_together | honest takes on AI
127.0.0.1
Joined September 2020
- Tweets37.7K
- Following4.9K
- Followers70.2K
- Likes57.4K
Wise retweeted
imagine building an AI agent that can shop for you, only to discover the store has its own agent
this is exactly how the agent economy is playing out.
> Yelp blocked outside agents and is now building it own internal agent
> it's no longer humans vs AI.
> it's whether your AI can access the platforms it needs.
Wise retweeted
I had a coding agent build a bike repair booking site in ~12 min,
and then it casually admitted it never actually clicked through the booking flow
so i gave TestSprite the URL and a quick spec, and this what happened:
> 4 agents poked around the site on their own
> wrote 21 end-to-end tests and ran them in real browsers
> checked closed sundays, slot lengths and cancellations
AI agents are good at writing code, but you still need something that tests it like a real user
Wise retweeted
this is why opus 5.5 is the best model in the world right now
it rewrote microsoft's typescript compiler in rust
about $24,000 in api usage
> an earlier openai attempt burned $400k+ and stalled at 84%
> the rust version is reportedly 2× faster than microsoft's own go rewrite
no one has read a line of the generated code
Wise retweeted
notice..
> Delta is blocking AI agents from booking flights
> JPMorgan Chase and Charles Schwab block agents from accessing accounts
> Yelp requires paid data licensing for agents
> Amazon has blocked Grok bot and other shopping agents.
the internet is starting to block AI agents that were supposed to do everything for us
Wise retweeted
here's how scope works and why it's such a big deal
1. it sits on your site and picks out which visitors are agents, not people
2. it tracks every agent from the moment it lands
3. you watch it live: intent, clicks, pages, errors
4. it tracks whether that agent converted and what it was worth
5. you click the issue where it failed and ship the fix
this is analytics rebuilt from scratch for ai agents
🤝 Paid partnership
We just solved attribution for AI Agents.
Introducing Journeys.
AI agents are visiting your product. Now you can see what they’re trying to do, what converts, and what’s making them churn.
With Journeys, you can:
- See the moment AI agents visit your product
- Follow the paths they take
- Track purchases, installs, and completed setup
- Pinpoint where agents get stuck and whats blocking them
- Review fixes prepared for you and publish them with a click
- Watch how those changes improve conversion
Turn agent traffic into revenue.
Try Journeys at tryscope.com
Wise retweeted
DigitalOcean just fixed the most annoying thing about running agents in prod
agent fails a user, logs look fine, you later find out 400 users we affected
> signals flags it in minutes
> simulations tests the fix before it ships
> a human approves, and the bug dies at user one
all in preview
🤝 Paid partnership
Wise retweeted
soon, most users of your product won't be people, they'll be AI agents
if an agent can't sign up and use your product on its own, a human has to step in and do it for them.
most people won't bother, they'll just pick another tool
that's what Restless is built for:
> it logs every API call, so you can see exactly where agents fail
> when a call fails, the error tells the agent how to fix it, with a link to the right docs
> if an agent gets stuck anyway, it can report what went wrong, and that becomes a ticket or a docs update for your team
so the more agents use your API, the more problems you find and fix. it keeps getting easier for the next one
🤝 Paid partnership
Wise retweeted
we keep asking which AI model is the smartest.
maybe we should start asking which one can actually run on the devices people already own.
this release makes a pretty good case for that conversation.
🚀 Introducing MiniCPM-V 4.7, our new open-source MLLM series!
Advancing the Performance–Efficiency Pareto Frontier.
It brings strong multimodal capabilities to iOS and Android devices with extreme efficiency.
Two new models:
🔹 MiniCPM-V 4.7 35B-A3B — Bring SOTA performance to local devices with newly introduced LES (Lookahead Expert Scheduling) technique, a new milestone of on-device multimodal AI.
🔹 MiniCPM-V 4.7 1B — Lightweight, efficient multimodal AI. Easier to deploy.
⚡ What's new?
SOTA performance: Competitive results across 34 benchmarks spanning multimodal understanding, STEM reasoning, OCR and long-video comprehension.
Fast on-device inference: LES predicts and preloads expert weights to reduce inference stalls, boosting 5× speedup on iPad Pro M4 (16GB) with byte-for-byte identical outputs. It enables MiniCPM-V 4.7 35B-A3B to achieve 17+ tokens/s on iPad Pro M4 (16GB) and 13+ tokens/s on iPhone 17 Pro Max, while costing only 3GB memory 🤯
Efficient visual processing: 4× and 16× hybrid visual token compression. Even at 16× compression, 35B-A3B scores 90.4 on OCRBench, 5.2 points ahead of Qwen3.6 35B-A3B.
Easier customization: We introduce Self-Replay Fine-Tuning, a simple way to customize new target capabilities at low cost while maintaining general performance.
Broad hardware support: Day-0 adaptations for Apple, Intel, AMD, Arm and Rockchip, with many more platforms supported through FlagOS!
Download, deploy, and build something we haven't imagined yet. 👀👇
🔗 GitHub: github.com/OpenBMB/MiniCPM-V
🤗 Hugging Face: huggingface.co/collections/o…
