标签：#c

You built a system to surface information so a person could decide. Somewhere it started deciding for them — the output stopped saying "here's what the documents show" and started saying "you should do X." Nobody designed that drift. An LLM, when asked a question, produces an answer-shaped thing, and an answer easily becomes a verdict. What everyone tries A prompt instruction: "Don't make recommendations." "Only state what's in the documents." People add the line and assume the boundary is enforced. Why it doesn't work A prompt instruction is a request, not a guardrail. The model follows it most of the time, then on the input that matters produces a confident recommendation anyway, because nothing structurally prevents it. "Please don't make recommendations" is to a guardrail what a sticky note saying "please don't enter" is to a locked door. And the stakes are higher than they look. When output drifts from evidence to verdict, accountability moves. As long as the system returns evidence and a human decides, the human owns the decision. The moment the system returns a verdict and the human defers, the system is deciding things it was never validated to decide — and when one is wrong, accountability is a blank. High-stakes fields separate evidence extraction from judgment on purpose; most RAG systems erase that line by default. The one shift Decide what the output is and enforce it structurally. An output should declare itself: answer, evidence, missing facts, or out-of-scope. "Return decision material, not a decision" has to live in the output contract and in gates — not in a polite request to the model. The system supplies frames; the human supplies verdicts. This is the output boundary — one of three places production RAG dies. Read the full version on my blog , where this connects to the RAG Failure Diagnosis Kit for teams debugging production RAG.

Build a Real-Time Crypto Trading Dashboard with Python, WebSockets, and React Real-time data is the difference between catching a move and reading about it later. This tutorial walks through a minimal but complete stack: a Python WebSocket client pulling live prices, a simple signal generator, and a React frontend displaying everything. You will end up with a dashboard that shows live Binance prices, basic momentum signals, and auto-updates without polling. Why this stack Binance provides a clean WebSocket API for tickers and trades. Python handles the backend connection and lightweight analysis. React keeps the UI reactive and simple to extend. No heavy frameworks, no paid data feeds. Prerequisites Python 3.11+ Node 20+ A Binance API key (read-only is fine for prices) Step 1: Python price stream Install the client library: pip install python-binance pandas Create price_stream.py : import asyncio import json from binance import AsyncClient , BinanceSocketManager import pandas as pd from datetime import datetime async def main (): client = await AsyncClient . create () bm = BinanceSocketManager ( client ) ts = bm . trade_socket ( ' BTCUSDT ' ) async with ts as tscm : while True : res = await tscm . recv () price = float ( res [ ' p ' ]) qty = float ( res [ ' q ' ]) ts = datetime . fromtimestamp ( res [ ' T ' ] / 1000 ) print ( f " { ts } | BTCUSDT { price : . 2 f } | { qty : . 4 f } BTC " ) if __name__ == " __main__ " : asyncio . run ( main ()) Run it: python price_stream.py You should see a live feed of trades. Keep this running as your data source. Step 2: Add a simple momentum signal Extend the script to calculate a 20-trade rolling average and flag when price deviates more than 0.3%: # inside the loop, after parsing res prices . append ( price ) if len ( prices ) > 20 : prices . pop ( 0 ) avg = sum ( prices ) / len ( prices ) deviation = ( price - avg ) / avg * 100 if abs ( deviation ) > 0.3 : print ( f " ⚡ Signal: { deviation : + . 2 f } % from 20-trade avg " )

👋 Hey dev.to community, Last week I shipped a workflow I built for a friend who runs an online shop. He called me again a few days later with a new headache: he's drowning in Purchase Orders. Every single one gets opened by hand, the data typed into a Google Sheet, and that sheet uploaded into his ERP to update his numbers. Hours a week, pure copy-paste. So I built him something to kill that step. He uploads the PO PDFs through a simple n8n form, and a structured Google Sheet comes out the other end. He just downloads it and pushes it to his ERP. How it's set up: The form accepts multiple PDFs at once , so he can batch a whole stack instead of doing them one by one. Each PO loops through on its own so nothing gets jumbled. The extraction runs on the easybits Extractor node ( @easybits/n8n-nodes-extractor ). I set the field structure up in two parts: the header fields that appear once per PO (PO number, PO date, delivery date, mark for, PR number, reference no), plus an articles array for the line items, each holding article name, unit and quantity. That array is the key bit, it gives you one entry per row of the PO table, and I flatten it into one sheet row per article with the header details repeated on each. Two things I added because real documents are messy: Error flagging . If any field comes back empty, the completion screen lists which document and which field didn't extract cleanly, so he knows exactly which PO to double-check instead of trusting it blindly. Document name column . The original filename lands in the sheet next to every row, so if a number looks off he can jump straight back to the source PDF. Workflow JSON is on GitHub: https://github.com/felix-sattler-easybits/n8n-workflows/blob/c38749a68fd6ea4ae6ebff41789d35cceaacdef1/easybits-purchase-order-extractor-workflow/easybits_purchase_order_extractor_workflow.json Anyone else automating document-to-sheet data entry? Curious how you're handling the messy multi-line rows – that was the trickiest par

The conversation around Agentic AI often focuses on one goal: making agents more autonomous. More tools. More reasoning. More planning. More independence. It sounds like progress. But is more autonomy always the right answer? As software engineers, we rarely optimize for "more." We don't build distributed systems when a monolith is sufficient. We don't introduce microservices because they're fashionable. We choose architectures that balance capability with complexity. The same principle applies to AI agents. The question isn't "How autonomous can my agent be?" It's "How autonomous should my agent be?" Autonomy Is a Design Decision When people talk about autonomy, they often think of it as a feature that an agent either has or doesn't have. In reality, autonomy is a design decision. Every time we allow an agent to make another decision on its own, we are increasing its responsibility. That responsibility comes with benefits, but it also introduces new engineering challenges. More autonomy means the agent can adapt to situations that weren't anticipated during development. It can make progress toward a goal without being guided through every step. At the same time, it becomes harder to predict, validate, debug, and trust. Autonomy isn't free. Thinking in Terms of an Autonomy Spectrum Instead of treating autonomy as a binary concept, it helps to think of it as a spectrum. At one end are systems that simply generate responses. They have no authority to take action. As autonomy increases, agents begin suggesting actions, invoking tools, planning multiple steps, and eventually deciding how to achieve a goal with minimal human involvement. The important observation is that every step along this spectrum increases both capability and complexity. That's why the objective shouldn't be to reach the highest level. It should be to stop at the level your problem actually requires. More Autonomy Isn't Always Better Imagine building an internal HR assistant. Its primary responsibil

There's a lot of absurdity in this industry right now. I'm curious if anyone else who has the ability to do so is quitting this month. If so I'm curious: 1. What pushed you to do it? 2. What will you be doing? (Even if nothing!)

I scraped every Show HN from the last 12 months (41,301 posts) plus the full comment tree of every launch with 10+ comments, ~100k comment timestamps, all from the Algolia HN API. The median launch gets 2 points and 0 comments. For launches that do get traction, half the comments they'll ever get arrive within 7.2 hours and 90% within 26, and the top decile decays on the same clock as everyone else. Vote timestamps aren't public, so comment timing is the attention proxy; caveats are in the post.

When going abroad, the right plugs are essential to keep your gadgets charged. These are my favorite travel adapters and chargers.

A company shifted from project- to product-thinking after their platform outgrew single-team use. The limitations that they felt with their platform were one-off deliveries, lack of product vision, and weak feedback loops. They have moved toward a self-service, API-driven, multi-tenant infrastructure with clearer ownership and better abstractions. By Ben Linders

Replay QA tells you what is broken before your users do Discussion | Link

There are now enough Amazon Leo satellites to launch the service, but its constellation is still tiny compared to Starlink's.