hensq

14 posts

hensq banner
hensq

hensq

@hensqcom

I build AI agents for real business workflows. I share the tests, failures, and results.

Katılım Kasım 2024
108 Takip Edilen4 Takipçiler
hensq
hensq@hensqcom·
@FreshworksInc The useful boundary in the case study is four bounded skills—password reset, account unlock, distribution lists, phone extensions—handling ~35% of IT volume while monthly tickets fell 39%. What time window and repeat-contact/escalation rate sit behind the 39%?
English
0
0
0
2
Freshworks Inc
Freshworks Inc@FreshworksInc·
iQor needed a solution that could: handle high ticket volumes, integrate with their in-house systems, and be AI-powered. With Freshservice, they built Jenny, an AI agent that handles password resets and account unlocks on its own. The result? Their monthly ticket volume dropped 39% in XYZ. Read the full story to see what worked for them: freshworks.com/customers/stor…
English
1
0
0
92
hensq
hensq@hensqcom·
@calcsam One boundary I’d test: a gate can prove the expected tool-call condition was met; it can’t prove the business decision was correct. In my local deterministic intake test, 30/30 frozen synthetic cases matched expected CONTINUE/STOP/HANDOFF across two runs. No LLM involved.
English
1
0
0
24
Sam Bhagwat
Sam Bhagwat@calcsam·
Today we're launching a major improvement to Mastra evals: gates and verdicts. Gates are binary checks that return 0 or 1, great for checking tool calls. Verdicts let you write `assert` tests to pass/fail in CI.
English
4
10
54
4.6K
hensq
hensq@hensqcom·
@LangChain Cross-agent keys help, but my publishing run hit three failures outside the agent: file picker, raw CDP, OS clipboard. Can LangSmith attach those external step outcomes to the same run tree without custom instrumentation, or does tracing end at the coding-agent boundary?
English
0
0
0
35
LangChain
LangChain@LangChain·
Cursor, Copilot, Pi, and OpenCode tracing: Now in LangSmith. Full session observability, no extra instrumentation. ✅Identify, group, and query any coding-agent trace with the same stable keys, regardless of which agent produced it ✅See the full run tree: turns, model calls, tools, and subagents ✅Token usage and cost per session, out of the box Blog by @harisaiharish langchain.com/blog/your-codi…
English
6
13
52
18.1K
hensq
hensq@hensqcom·
@LangChain Before enabling this, I ran a preflight: my local Codex package is 0.117; the docs require 0.128+. My bigger question is data scope—the plugin sends full transcripts and tool I/O to LangSmith. Is there a supported redaction layer before traces leave the machine?
English
0
0
0
63
LangChain
LangChain@LangChain·
We built a tracing plugin for every Codex session into LangSmith. Now every turn (tool calls, token usage, subagent threads) lands in LangSmith as a real trace you can dig into. Two config blocks and one flag, and it's live.
English
7
7
46
8.1K
hensq
hensq@hensqcom·
I tested the test. It failed. Four required fields were hard-coded null, so READY was unreachable. I fixed the local gate, then ran 30 cases frozen before coding. All 30 matched the expected CONTINUE, STOP, or HANDOFF decision. No LLM, customer data, or price calculation.
hensq tweet media
English
0
0
0
25
hensq
hensq@hensqcom·
@aidan_mclau Codex is the easy part. I’d call an agent a research intern only if every claim points to a source, conflicting evidence stays visible, and another researcher can retrace the evidence. Otherwise it’s still a fluent output machine.
English
0
0
0
334
Aidan McLaughlin
Aidan McLaughlin@aidan_mclau·
"when will get the automated research intern?" idk are we talking about a research intern who can use codex
English
16
3
348
17.1K
hensq
hensq@hensqcom·
@ComfyUI @PurzBeats @jojodecayz @mattmiller_ai Letting the agent edit the graph is only half the system. The other half is making every run debuggable: persist the submitted workflow JSON, inputs, job ID, and any node-level errors returned. Otherwise you can't tell a bad graph edit from a runtime failure.
English
1
0
0
31
ComfyUI
ComfyUI@ComfyUI·
Livestream Alert: Run ComfyUI From Claude/Cursor with Comfy MCP Host: @PurzBeats Comfy MCP lets Claude, Cursor, Amp and almost any AI agent you're already using build, run, and iterate real Comfy Cloud workflows for you. Join @jojodecayz, & @mattmiller_ai who's been building Comfy MCP from day one, for a live walkthrough, plus a first look at what's coming next. In this stream: → What Comfy MCP is and how to connect your agent → A live look at an agent building, running, and editing a real workflow through MCP → Live Q&A look at the future of Comfy MCP →Live Q&A Click the link below to tune in👇
English
6
12
63
7.1K
hensq
hensq@hensqcom·
@jxnlco Three models in one AWS stack make model choice easier. The harder question is routing by the cost of a wrong answer. Reversible tasks can use a cheaper model; high-impact cases should move to a stronger model or human review under one escalation policy.
English
0
0
0
14
jason
jason@jxnlco·
GPT-5.6 Sol, Terra, and Luna are now available on Amazon Bedrock. Use the Responses API in AWS, with first-party pricing and usage that counts toward AWS commitments. Teams can also configure GPT-5.6 for ChatGPT Work and Codex. aws.amazon.com/blogs/machine-…
English
8
3
87
7K
hensq
hensq@hensqcom·
@llama_index @TauriaP The bounding-box preview is the key feature here. Clean Markdown can hide a bad extraction; source boxes make it reviewable. Before an agent acts on the result, I’d still validate required fields and route missing or low-confidence cases to a human.
English
0
0
0
39
LlamaIndex 🦙
LlamaIndex 🦙@llama_index·
Ever wanted to quickly turn a PDF into clean text to paste into your favorite AI agent, without having use CLIs or open the browser? We built exactly that. Using @TauriAp ps, with a Rust backend powered by LiteParse and a React frontend, we created a cross-platform desktop app that lets you: • Drag and drop PDFs • Convert them into clean Markdown • Preview page screenshots with extracted bounding boxes overlaid for easy inspection It's a simple way to understand your document's structure and verify the parsing results before handing them off to an LLM. Check out the demo below! 👇 GitHub: github.com/run-llama/lite… Get started with LiteParse: developers.llamaindex.ai/liteparse/
English
7
6
49
19.3K
hensq
hensq@hensqcom·
Lower inference cost opens more use cases. But enterprise adoption also depends on lowering verification cost. A cheap agent output is still expensive if a human must recheck every result. Evals, traceability, exception handling and rollback belong on the same cost curve.
Aaron Levie@levie

The best way you’re going to continue to get large scale agentic adoption is by continuing to bring down the cost of intelligence. More use-cases open up for AI every time you can have lower cost tokens (for the same or better level of capability). Almost all information work in the future will involve an agent somewhere in the workflow creating, processing, reviewing, or classifying data in some way. This will happen sooner *or* later depending on the cost of tokens of frontier models. Whether this happens from closed or open models is somewhat incidental, but the key is just that it happens. It’s great to see so much innovation and different approaches in AI right now as there are so many more use cases to power.

English
0
0
0
76
hensq
hensq@hensqcom·
账号介绍： 记录AI 数字员工落地实践-build in public 主要记录 3 类内容： 1. 项目实践：如何从 Excel、文档、群聊、老员工经验和人工判断，慢慢做成可运行的 AI 系统； 2. 企业落地：岗位职责梳理、业务知识提炼、心态变化应对、踩坑记录等； 3. AI技术学习：skill设计、agent设计、工具使用等。
中文
0
0
0
40
hensq
hensq@hensqcom·
Claude Opus 4.7 Qwen3.6-35B-A3B Gemini MacOS App Interesting！
Română
0
0
0
96
hensq
hensq@hensqcom·
Claude code This isn't working right now. You can try again later. 500 error，what happen？
English
2
0
2
643

Keşfet

@FreshworksInc @calcsam @LangChain @harisaiharish @aidan_mclau @ComfyUI @PurzBeats @jojodecayz