Seth Vargo

Structured Events in Rails! OMG! I've been waiting for this for ~4 years. Want to send *proper* structured events? Yeah. Me too. This is the most exciting development in observability for Rails in... ...well, forever. This opens the door to proper OpenTelemetry instrumented Rails apps... ...with a single subscriber and an intuitive interface. But it's way more than that - it allows us to instrument *business* events. Things like: * Order Processed * Customer Paid * Invoice Created I'm beyond excited about the direction the Rails team has gone in. * Decoupled the subscriber from the instrumenter * Aligned the API with the ridiculously good error reporting * Namespaces for events This will open the door to massive improvements in the observability of Rails apps. A HUGE thanks to Adrianna Chang for leading this effort and getting a huge feature into Rails that's critical for modern observability. Kudos! Comment below with your questions about how to use observability and I'll answer what I can.

11

6 Comments

Check out this a brief demo of a small tool I have been building in the last couple weeks that turns dense federal appropriations PDFs into something regular folks can understand. Right now the pipeline looks like this: • A Next.js frontend uploads bills via presigned URLs straight into MinIO (S3-compatible storage). A FastAPI backend pulls the PDF, runs a parser that merges table cells and surrounding prose into structured rows, attaches page and bounding box metadata, and stores everything in PostgreSQL. • It then uses the OpenAI API to suggest cleaner program names and short summaries for each funding line. • From there the API can audit the table, generate diffs between fiscal years, export CSVs, and render highlighted snippets of the original PDF for context. Everything runs under Docker Compose: Postgres for relational data, MinIO for documents and images, FastAPI for the API, and Nginx as the single public entry point. The goal is to make it much easier to see what is actually being funded, compare versions of a bill, and link every number back to the source text. It's also a pretty neat use-case for use LLMs to create accessible tools that benefit everyone and break down barriers. Check out the code here: https://lnkd.in/eiQDzBuX

2

It’s worth being clear about why current tools do not solve the issues with identity ownership and replication that I'm working on right now. A lot of people assume copyright, takedowns, or platform moderation will fill the gap. They won’t. Copyright protects works, not identity. Takedowns are slow, reactive, and jurisdiction-dependent. Platform rules only apply inside walled gardens, and often work under severely misaligned incentives. Most importantly, none of these mechanisms operate at AI speed. You cannot moderate your way out of an infrastructure problem. You have to design the right primitives upstream. #AIInfrastructure #ContentModeration #DigitalIdentity #TechDesign

6

Why is tracking government spending still this hard in 2026? Every state runs its own spending websites. Different formats. Different layouts. No shared APIs. No standards. Some states even have 5 or more separate spending portals; none designed for programmatic access. States that do have APIs are using different solutions, with different fields and different authentication methods. For example:   • NY, TX, IL, WA, HI, CO, MD, CT, and others use Socrata   • CA uses CKAN I pushed this as far as I could:   • Built scrapers for some states that currently don’t have an API   • Built rules-based checks   • Ran Benford’s Law deviation analysis   • Tested anomaly detection   • Used all the results to calculate a risk score And the result? Inconclusive. Not because the methods don’t work, but because the data is incomplete, inconsistent, and hard to access. Real impact of this mess:   • Cross-state analysis is painful   • Fraud detection doesn’t scale   • Transparency tools constantly break   • Citizens and watchdogs hit unnecessary walls The fix is not complicated: 👉 Standardize state spending data APIs, like we already did at the federal level with USAspending.gov. With real data access:   • Automated fraud detection actually works   • AI becomes useful instead of hand-wavy   • Accountability improves We solved this once with the DATA Act. State-level spending should be next. Would love to hear from folks in govtech, data, or policy who’ve hit the same wall.

1

During the "Ambient Agents" meetup last week, Brace Sproul gave us a walkthrough of Open SWE, a coding agent built on Langgraph. Open SWE monitors GitHub issues. When it gets a new one assigned, it kicks off a multi-agent architecture with a Manager, Planner, and Programmer. It can receive messages from users at any time during the process and incorporate them into the state. Check out the slides for diagrams explaining each agent: https://lnkd.in/g3PnNHqG And the code: https://lnkd.in/gY68TKYy I found the takeaways really interesting: * Summarization is hard: Especially with code changes that can get quite long, you literally can't send the entire context history to the LLM, and highly detailed old diffs may distract it. IIRC, the agent sends the last 20 messages, and summarizes the changes before. * Agentic > hard coded: Open SWE tends to use the agentic approach of "here are your tools, pick which one to use" for almost all tasks. Except for one.. * Explicit error handling: When an error happens, Open SWE stops everything and focuses on resolving the error. Otherwise, it could get too far afield and be unable to get back to a good place to solve the error. (Hopefully I didn't mis-represent anything I heard!) #Langchain #Langgraph

93

3 Comments

"Overconsumption is Killing Open Source This isn't some random complaint from a few disgruntled maintainers. We're talking about an unprecedented joint statement from the stewards of virtually every major package repository - Maven Central, PyPI, npm, RubyGems, The Rust Foundation, the Eclipse Foundation, and others." "The coalition's proposed solutions are reasonable but firm. High-volume commercial users should contribute financially through partnerships or tiered access models. Companies need to implement better caching and reduce wasteful usage." https://lnkd.in/ehPAe_Mw

108

3 Comments

🕰️ Philly Tech First #15 Before software platforms. Before distributed systems. Before “network effects” was a phrase. In 1846, the Pennsylvania Railroad was chartered and headquartered in Philadelphia. By the late 1800s, it had become the largest railroad in the world — and one of the largest corporations on the planet. But this wasn’t just about trains. It was about systems engineering at scale. The Pennsylvania Railroad pioneered: → Standardized time zones and scheduling coordination → Integrated telegraph-based dispatch systems → Formal corporate management hierarchies → Operational safety protocols across a vast network This was one of the first truly complex, multi-state, real-time operational networks in American history. Steel and steam — yes. But also: Data. Coordination. Infrastructure logic. Philly didn’t just lay track. It built one of the earliest large-scale networked systems in the country. That’s technology. #PhillyBuilds

25

1 Comment

I’ve been reviewing a case involving a software development company operating under the name Trango Tech, and it raised some questions I’d like to put out to the community. This isn’t a complaint and it’s not an accusation — it’s about understanding whether certain experiences are isolated or shared. In this case, a business engaged a development firm that presented itself as having an international footprint (USA / Australia), a sizeable team, and experience delivering complex projects. Contracts were signed for an app and supporting web platforms. What followed was interesting: • A substantial amount of work was completed • The project reached a high percentage of completion • Delivery timelines extended well beyond what was originally agreed • Bugs and issues were acknowledged but took months to resolve • The project ultimately stalled before being fully completed or handed over On paper, this looks like a project that was almost finished — yet never quite crossed the line. I’m not drawing conclusions. I’m simply trying to understand whether this type of experience is unique, or whether others have encountered something similar when working with the same company. If you’ve had dealings — positive or negative — and are comfortable sharing information privately, feel free to message me directly. I’m particularly interested in learning whether others experienced: • Projects that were “nearly done” but never completed • Extended delivery timelines • Ongoing fixes without resolution • Unclear handover or IP outcomes This post is about listening, not blaming. If there’s no broader pattern, that will become clear too. #SoftwareHouseLeaks #DannyDeHek #TrangoTech

34

25 Comments