Yizhuo Wang

I wrote an article on how we translate user-generated content at scale at Mercari. It summarizes two years of work with various models, LLMs but not only, and how cost evolved, reducing it 100x over the period. I also go over non-AI features and user experience. https://lnkd.in/gbpQAQmX

39

2 Comments

Localization is more than translation. Pendo.io just rolled out our first localized experience—in Japanese 🇯🇵—designed for Japan’s fast-growing, AI-driven tech market. Go Pendo.io Japan ! From login screens to product areas, we’re bringing product-led growth to users in their language. Learn how we're doing it—and why Japan was first! 👇

2

An agent has no consistent definition. You can think of an agent as an autonomous or semi-autonomous system that can take actions on behalf of the user in a given environment, state, and make decisions or take actions that can accomplish certain tasks along a given time-frame, either by calling tools, dynamically selecting the next action to take or use a deterministic "workflow" (systems where LLMs and tools are orchestrated through predefined code paths). Using an Agent framework that gives you the ability to built systems in both, a reliable and a dynamic way can accelerate your agent development journey using agent abstractions. This means building Agentic systems that can reliably call tools, store memory (episodic/semantic/procedural), have comprehensive logging and observability, human in the loop workflows and the ability to build various multi-agent patterns flexibly based on your use case. A successful Agentic system in production is usually a combination of both, dynamic and predictable/reliable multi-agent systems. Strands Agents SDK gives you exactly these capabilities by treating each “agent” as a combination of a foundation model plus a suite of tools. You define a prompt and register your tools (decorated functions) in code, then Strands handles reasoning→planning→tool-execution cycles, local testing, and cloud deployment (ECS, Fargate, Lambda, EC2), along with support for all other agent abstractions provided above. Excited to share a hedge-fund analyst multi agent system: This uses the newest Anthropic's Claude 4 Sonnet/Opus that powers the Lead Analyst Agent, routing incoming queries to specialized sub-agents for fundamental, technical, and market analyses. Each specialist is wrapped as a callable tool (using the “agents-as-tools” multi-agent pattern), so the orchestrator never has to implement domain logic itself and can handoff the task to an agent as a tool. For sensitive operations (insider lookups), we utilize a HITL approval step that halts execution until a human grants consent. We also use meta-tooling that enables the Lead Analyst to generate, load, and invoke new custom tools at runtime—whether it’s a portfolio beta calculator or a pricer—without redeployment. Strands also embeds observability (Langfuse) and OpenTelemetry tracing so you can trace reasoning events, tool invocations, errors, and end-to-end workflows in real time. View more information on the code implementation here: https://lnkd.in/gJmwVyGi Code implementation: https://lnkd.in/gzTtJvJq Thanks to 🏄‍♂️ Cagatay Cali for being a reviewer/collaborator on this! Feel free to try it out and reach out with any questions/ideas. #aws #agenticAI #strands #agents #generativeAI

60

6 Comments

Agent Ready Launch Week Day 2 - Native Streamable HTTP MCP Server https://lnkd.in/g8Fzmqbp Today I'm excited to talk about our new native Streamable HTTP Model Context Server with Context Aware Tools! Our new native MCP server makes it easy to build multi-tenant agents or access your knowledgebases from existing MCP clients by simply connecting to a Streamable HTTP endpoint.  Deployment becomes easy without needing to worry about installing npm packages or managing processes from a stdio server. And while easy deployment is great, here's what makes it special: Ragie's MCP server is smart. As content gets added to your knowledgebase, Ragie continually updates the tools' descriptions so when your MCP client needs to decide which tool to use, it has the right context to make a good decision.  This minimizes the amount of bad tool choices, hacks and explicit instructions needed to force clients to use a tool. No more prompt engineering just to get your tools to work. With Context-Aware Tools, clients have all of the context they need.

12

You can now ask ChatGPT to handle requests like “look at my calendar and brief me on upcoming client meetings based on recent news,” “plan and buy ingredients to make Japanese breakfast for four,” and “analyze three competitors and create a slide deck.” ChatGPT will intelligently navigate websites, filter results, prompt you to log in securely when needed, run code, conduct analysis, and even deliver editable slideshows and spreadsheets that summarize its findings. https://lnkd.in/gTsXSQYU

15

Very excited about Kong Inc. extending "API governance" abilities to eventing systems. Our new event_gateway can front Kafka brokers and let you run policies to mutate messages. The event_gateway integrates back into Kong Konnect - our SaaS API platform - one single platform to "build, run, govern" http and eventing apis. ps: built in Rust on tokio, the performance characteristics of the event_gateway are amazing.

77

3 Comments

There’s been a lot of great discussion lately about what it truly means to become an AI-native organization. At Mercari, we’ve been on that journey too — and one thing I’ve learned along the way is that transformation doesn’t start with technology. It starts with understanding yourself. For us, that meant getting real about our data. Using DX, we built custom dashboards to track patterns, measure impact, and uncover what was actually changing — turning what used to be gut feelings into clear, measurable insights. The DX team partnered closely with us, helping us connect the dots (and the data! 🤩) so we could finally see how AI was reshaping the way we write, review, and ship code. We were not only able to debug our workflows — but even gain visibility into our organizational dynamics, which was truly eye-opening! I’ve come to realize that the best tools don’t just measure progress — they provide clarity. And clarity? That’s what drives real speed. #MOVEFAST 🚀 https://lnkd.in/gw9T3aCx

39

PMs, Engineers, and English as Code? ----- In the era of AI, is English and programming code really that different? Karl Chan, founder of Sourcy, and I hit upon this question last night about the future of how PMs will work with engineers. . 1. Obvious Differences ----- > English as we understand can be non-deterministic – two people hearing the same “vague” thing can have different interpretations. > We think of code differently however. When we think of code, we think of the same inputs as always producing the same outputs. > (This isn’t always technically true, but more on that later). . . 2. Specs as Code ----- > But what if highly precise words in English were used to convey a process in a spec – with agentic coding agents then executing code writes based on that spec? > To the extent that you can expect/assume highly compliant code changes, is this then not a “programming language” of sorts? . . 3. Deterministic LLMs? ----- > This is especially in light of Thinking Machine’s recently published paper titled “Defeating Nondeterminism in LLM Inference”. > Thinking Machine achieved 100% of the same results running a prompt 1000 times using experimental inference-stage architectural changes.  > For the layperson, this means we may actually be quite close to exact LLM outputs for a given input, just like a program! > (FYI: Thinking Machines is an AI startup founded by Mira Murati, the former CTO of OpenAI) . . 4. Rethinking Programming ----- > If today, we accept that LLMs do a lot of a program’s heavy lifting, and > That traditional coding is just the scaffolding to control and guide this power > Then the expected value of programming cannot be said to be simply a deterministic output > And following that logic, if we can get such expected value from writing (and collaborating on) specs and have coding agents do the tedious “coding”, then > Is English – in the precise form used in writing specs – not therefore a programming language? . . 5. Spec-English as a Language ----- > After all, traditional programming languages have a wide range of abstractions themselves to begin with.  > We have low level programming languages like C all the way to high level programming languages like Python. > Perhaps we should see Spec-English (I wonder if this can become a trending term) as the next highest level programming language. . . 6. The Future of PM + Engineering ------ > We really might be in an era where PMs and engineers focus all their time on debating and refining specs, with the (eventual) full confidence and trust that coding agents will be able to execute that spec flawlessly. > Imagine how fast products can get developed and iterated on.   User gave feedback? Boom – changes/features were incorporated into production within a day. > What a world this would be!

15

2 Comments