Clay Wolcott

[Clay Wolcott @CSWolcott]

@InternetH0F Vozinha took this photo!

[Clay Wolcott @CSWolcott]

@andrewamann On the waitlist!

[Clay Wolcott @CSWolcott]

Utilized @mvanhorn 's /last30days to get up to speed on the latest in AI data ingestion best practices. Thank you Matt!

What I learned:

Ingestion is the real bottleneck, not the model -- The strongest signal across X: "The RAG problem was actually an ingestion problem," per @0xlelouch_. Engineers are being judged on pipeline quality, not on whether they can recite transformer variants. @michentr put it bluntly: "RAG is only as good as your data ingestion pipeline." The web sources agree -- most failed deployments nail chunking/retrieval but skip data quality and monitoring.

Curation beats volume -- 500 well-governed docs consistently outperform 50,000 poorly governed ones (per Data Engineer Academy). Every document needs an owner responsible for freshness, removal, and metadata tagging. This is the single highest-leverage practice and the one most teams skip.

Self-updating knowledge bases are the emerging pattern -- "The traditional corporate wiki is officially dead. Teams are shifting to agentic (self-updating) knowledge bases," per @femke_plantinga (711 likes, the top post). Data that maintains itself instead of sitting stale.

Multi-provider / heterogeneous pipelines are maturing -- Teams moving past single-API setups to pipelines that route across providers and handle mixed formats, per @CalebMccombs2. Docling + Ray on OpenShift for scaling document parsing was a concrete stack shared by @michentr.

KEY PATTERNS from the research (the actual best practices):

-- Parse before you chunk -- Strip boilerplate, nav, disclaimers first. Use Unstructured, Apache Tika, or OCR for scanned PDFs. Garbage in the parse stage poisons everything downstream, per Databricks.

-- Start simple on chunking -- Recursive or fixed-size (~512 tokens, 10-20% / ~50-token overlap) is the recommended default. Semantic chunking works but adds LLM-call cost. Don't over-engineer early, per Redis/Weaviate.

-- Attach metadata to every chunk -- Topic, date, source, ownership. Retrieval filters and ranks on this. Non-negotiable for accuracy at scale.

-- Instrument the last two stages -- Groundedness rate, hallucination regression thresholds, plus latency/cost. Failed deployments skip continuous eval and monitoring, per Data Engineer Academy.

-- Retrieval-native access control -- Filter each result by user identity and document-level policy before augmentation. The 2026 security default for sensitive data, per Kiteworks/Informatica.

✅ All agents reported back! ├─ 🔵 X: 26 posts │ ~1,180 likes │ ~150 reposts ├─ 🌐 Web: 8 pages -- Data Engineer Academy, Databricks, Kiteworks, Informatica, Redis, Weaviate, https://t.co/ClXzzuRhIO, Microsoft Learn └─ 🗣️ Top voices: @femke_plantinga (711 likes), @0xlelouch_, @michentr, @CalebMccombs2

Note: Reddit and Hacker News APIs errored out this run, so this is weighted toward X and web sources.

Skill for reference: https://t.co/izeNmf1nw9

[Clay Wolcott @CSWolcott]

@jpshrodinger Interesting area for sure. Openclaw reminding me to refill the water filter every couple nights kept our water levels consistently high. Good result

[Clay Wolcott @CSWolcott]

@DanKornas What’re integrations with this like. Will it work with Nest, Ring, Roomba, Temu motorized blind company?

[Clay Wolcott @CSWolcott]

I don’t get it. Who do I see at MIT about getting one of these licenses

[Clay Wolcott @CSWolcott]

I tried building an autonomous agent to collect plumbing repair quotes. Building out end-to-end automation for a one-time use case is impractical, but I see this is a valuable task for an autonomous house manager.

Goal: save $200 per hour spent getting additional quotes

Background: There’s a separation in the drain line for the half-bath toilet that will cost $4,040 to fix

At 2 hours, this is where we are:

1. The automation is a Claude skill on a scheduled task with a Gmail connector

2. I fed Claude the PDF of the plumbing inspection and instructed it to create a statement of work (SOW) for exactly what needs to be done and nothing more. We were going to email vendors and handle all correspondence through that channel.

3. Claude created the SOW.

4. Claude created a vendor outreach tracker. I had three referrals and Claude found 4 through online research.

5. Claude drafted an initial outreach to the 7 vendors for my review. There was no variability between them so I approved one and I sent the rest.

6. Claude then monitored my inbox every hour, updated the tracker with any new correspondence, and drafted a reply. I'd receive an audible ping when it needed my attention.

Two things quickly became clear: 1) Vendors want to do a site inspection to generate an accurate quote. Each inspection will require some of my time, so polling a large list of vendors no longer feels reasonable. 2) Vendors want to talk on the phone. A voice agent could handle this if I take one phone conversation and talk about anything and everything to generate as many Q&A’s as possible for the agent.

This project won’t get the full end-to-end automation treatment with voice agent and automatic scheduling, but it proved to be a good testing ground for the next home improvement project.

Important items to have for the next:
- SOW
- Inspection availability (google calendar appointments)
- Voice agent loaded with an FAQ and google phone number

[Clay Wolcott @CSWolcott]

@usmntonly Correction:

Round of 32 — Bosnia
🇧🇪 Round of 16 — Belgium
🇪🇸 Quarterfinal ��� Spain
🇫🇷 Semifinal — France
🏴‍☠️Final — Cabo Verde

[Clay Wolcott @CSWolcott]

@AlexFinn So much bullshit. Everyone’s going to microcenter and getting deal of a lifetime. Dummies

[Clay Wolcott @CSWolcott]

@jun_song The 48gb unified memory has been $5k for months. This is fake. Price increases haven’t hit existing product

[Clay Wolcott @CSWolcott]

@AndrewBolis YT

[Clay Wolcott @CSWolcott]

@ThisWeeknAI I’m in

[Clay Wolcott @CSWolcott]

@Jason I’m in

[andrew chen @andrewchen]

prediction re the end of spreadsheets

AI code gen means that anything that is currently modeled as a spreadsheet is better modeled in code. You get all the advantages of software - libraries, open source, AI, all the complexity and expressiveness.

think about what spreadsheets actually are: they're business logic that's trapped in a grid. Pricing models, financial forecasts, inventory trackers, marketing attribution - these are all fundamentally *programs* that we've been writing in the worst possible IDE. No version control, no testing, no modularity. Just a fragile web of cell references that breaks when someone inserts a row.

The only reason spreadsheets won is that the barrier to writing real software was too high. A finance analyst could learn =VLOOKUP in an afternoon but couldn't learn Python in a month. AI code gen flips that equation completely. Now the same analyst describes what they want in plain English, and gets a real application - with a database, a UI, error handling, the works. The marginal effort to go from "spreadsheet" to "software" just collapsed to near zero.

this is a massive unlock. There are ~1 billion spreadsheet users worldwide. Most of them are building janky software without realizing it. When even 10% of those use cases migrate to actual code, you get an explosion of new micro-applications that look nothing like traditional software. Internal tools that used to live in a shared Google Sheet now become real products. The "shadow IT" spreadsheet that runs half the company's operations finally gets proper infrastructure.

The interesting second-order effect: the spreadsheet was the great equalizer that let non-technical people build things. AI code gen is the *next* great equalizer, but the ceiling is 100x higher. We're about to see what happens when a billion knowledge workers can build real software.

[Clay Wolcott @CSWolcott]

@rbfasho Access por favor

[Dr. Parik Patel, BA, CFA, ACCA Esq. @ParikPatelCFA]

Never deleting this app

[Clay Wolcott @CSWolcott]

I'm helping test @friendtech, the marketplace for your friends on Base 🐰

Get the app and search for my Twitter username to be an early holder of my keys 🤗