Sign in

Matthew Mullins

@mmullins.coginiti.co
4K followers 543 following 1.9K posts

technologist, philosopher, and outdoors enthusiast CTO@COGINITI

PostsRepliesMedia
Matthew Mullins @mmullins.coginiti.co · 30/09/2026
For those wondering where DuckDB was going to show up in AWS products post DuckLabs acquisition, here is use case one. Apache Iceberg and Parquet queries in Aurora Postgres. #databs
aws.amazon.com
Amazon Aurora PostgreSQL now supports direct querying of Apache Iceberg and Parquet data in your data lake | Amazon Web Services
Amazon Aurora PostgreSQL now lets you directly query Apache Iceberg and Parquet data stored in your data lake alongside live operational data—no ETL pipelines required. Powered by DuckDB embedded with...
130
Matthew Mullins @mmullins.coginiti.co · 30/09/2026
I’ll never cease to be amazed how some dudes can make a data platform their entire personality.
000
Reposted by Matthew Mullins
All Things Open @allthingsopen.bsky.social · 27/09/2026
Matthew Mullins (@mmullins.coginiti.co), CTO for @coginiti.co, joins us at #AllThingsOpen to present "Beyond the Table: Data Strategy and Governance in the Age of Apache Iceberg"! 2026.allthingsopen.org/sessions/bey...
On one side is Matthew Mullins's headshot. Text on the other side reads Matthew Mullins. CTO, Coginiti. ATO 2026. October 19 - 20. Raleigh, NC USA
012
Matthew Mullins @mmullins.coginiti.co · 27/09/2026
Next Wednesday I’m talking with @jtalisman.bsky.social, author of The Ontology Pipeline, about the distinction between data management and knowledge management. We’ll talk ontologies, context, knowledge infrastructure, and what AI systems are still missing. 9/30 at 2:30 PM ET
riverside.com
Data Is Not Knowledge
040
Matthew Mullins @mmullins.coginiti.co · 27/09/2026
Hand grinding coffee beans in the dark and didn't notice I was missing the base to capture the grind. So off to that kind of day...
020
Matthew Mullins @mmullins.coginiti.co · 27/09/2026
Pretty much how I was taught to engage in philosophical disagreements.
020
Matthew Mullins @mmullins.coginiti.co · 23/09/2026
BigDataLDN off to a banger start! Standing room only for my talk on the evolution of semantic layers and why they matter for context, governance, and costs in the age of agents! #bigdataldn
031
Matthew Mullins @mmullins.coginiti.co · 17/09/2026
Weirdly felt the need to write this corrective on what counts as a semantic layer. #databs
fieldnotesondata.substack.com
The Semantic Layer Is Not an Ontology
A brief note on baptism, inheritance, and why technical terms have histories
040
Matthew Mullins @mmullins.coginiti.co · 16/09/2026
Telling a real semantic layer from a rebranded metrics catalogue, in 3 questions: 1. Can anything other than your BI tool read it? 2. Is policy enforced here or only documented here? 3. Does it survive a question nobody anticipated? #BigDataLDN #SemanticLayer
010
Matthew Mullins @mmullins.coginiti.co · 15/09/2026
I'm concerned about the rampant Chronic Intent Attribution Syndrome (CIAS) — the tendency to interpret a model’s output as evidence of private motives, elaborate plans, and a burning desire to fool the evaluation team — running amok in AI labs.
110
Matthew Mullins @mmullins.coginiti.co · 14/09/2026
"Can't we just tell the agent what it's allowed to see?" You can. It'll mostly kinda work, and "mostly" is doing a lot of work there. Prompt policy is a request. Semantic layer policy is a boundary. #BigDataLDN #DataGovernance
010
Matthew Mullins @mmullins.coginiti.co · 09/09/2026
An agent's confidence is not correlated with its correctness. When an analyst doesn't know how you define revenue, they ask, and the question is visible. When an agent doesn't know, it picks, and the picking is invisible. #BigDataLDN #AIAgents
000
Reposted by Matthew Mullins
coginiti.co @coginiti.co · 09/09/2026
Agents stopped failing at SQL. For the strongest model tested, 60%+ of remaining failures were semantic: wrong metric, wrong source, wrong join key. New research on what survives as models absorb the stack.
coginiti.co
Better Models Make Semantic Context Matter More | Coginiti Blog
A Berkeley/Stanford paper shows better models absorb agent scaffolding while semantic failures dominate. Why persistent, governed semantic context becomes the durable engineering problem.
011
Matthew Mullins @mmullins.coginiti.co · 01/09/2026
Wendell Berry is dead and I don’t feel so good myself. Peace of the Wild Things is the poem I’ve returned to most in life as a general comfort.
020
Matthew Mullins @mmullins.coginiti.co · 01/09/2026
Excited for the release of Coginiti 26.8. A highlight: Semantic Lineage, helping teams understand not just where data comes from, but how business meaning connects across entities, metrics, and transformations. Another step toward governed, semantic, AI-ready analytics. coginiti.co/blog/coginit...
coginiti.co
Coginiti 26.8 Release Notes | Coginiti Blog
Coginiti 26.8 deepens semantic modeling with a lineage graph, autocomplete and linting in the SMDL editor, and data quality testing for the Semantic Layer. It also adds synonyms,…
010
Matthew Mullins @mmullins.coginiti.co · 26/08/2026
Wrote up some thoughts on what the AWS acquisition of DuckLabs means for users and builders. (I think it's mostly good news). Congrats to Hannes, Mark, and the whole team at DuckLabs
fieldnotesondata.substack.com
What Changes When AWS Owns DuckLabs
AWS announced today that it has signed an agreement to acquire DuckLabs, the company behind DuckDB.
171
Reposted by Matthew Mullins
Michael Bench-Dracula @mikebenchcapon.bsky.social · 24/08/2026
1 The world is all that is the case 1.1 The world is made up of facts, not things 1.11 Facts don't care about your feelings 1.111 😢
1357
Matthew Mullins @mmullins.coginiti.co · 21/08/2026
We're moving data quality up stream to the semantic layer. Why? Because your tables can pass every data quality test and your business metrics can still be wrong. CoginitiScript tests now work with Semantic SQL, making it possible to test metrics as the business actually consumes them. #databs
coginiti.co
Testing Meaning, Not Just Rows: Data Quality at the Semantic Layer | Coginiti Blog
Row-level tests validate ingredients, not recipes. How CoginitiScript and Semantic SQL bring assertion-based testing to metrics, dimensions, and relationships.
030
Matthew Mullins @mmullins.coginiti.co · 20/08/2026
Some people are reading this article about Sen. Kevin Cramer and Trump nominees as a sign of GOP pushback on Trump. Nope, the real subject is a bipartisan issue about the future of North America’s “Duck Factory” which is a huge conservation story that's barely registered here on bluesky.
thehill.com
GOP senator places hold on Trump nominees over wetlands policy
Republican Sen. Kevin Cramer (N.D.) is placing a hold on all future Trump nominees over the administration’s approach to wetlands.  In a statement shared with The Hill, Cramer expressed disapp…
220
Matthew Mullins @mmullins.coginiti.co · 20/08/2026
Great evening out with the Wake Dems. @ross.house.gov was on fire with her remarks and generously took town hall questions. Lot of good energy in the room going into the midterms.
000
Matthew Mullins @mmullins.coginiti.co · 19/08/2026
Excited to be headed to #BigDataLDN next month and talking about semantic layer goodness. Come for the talk and swing by the Coginiti booth to chat all things data, semantics, and AI. Tickets are free, so no reason not to attend if you're in the London area. www.bigdataldn.com/en-gb/regist...
131
Matthew Mullins @mmullins.coginiti.co · 17/08/2026
A career in data takes more than learning the right tools. I’m talking with Aaron Wilkerson of Carhartt next Tuesday about what actually gets you to the next level: technical skills, business impact, leadership, career choices, and what AI changes. riverside.com/webinar/regi...
071
Matthew Mullins @mmullins.coginiti.co · 13/08/2026
Had the chance to meet @paulbarringer.com running to be North Carolina's 13th district representative. He's an a real Eagle Scout, AT trail hiker, grew up and lives on a farm, and most importantly understands the needs of North Carolina voters. Working for change in the midterm elections!
252
Matthew Mullins @mmullins.coginiti.co · 11/08/2026
Want to meet the next congressman for NC13? Meet and Greet with Paul Barringer at Norse Brewing in Wake Forest tomorrow.
mobilize.us
Meet and Greet with Paul Barringer at Norse Brewing in Wake Forest! · Paul Barringer for Congress
Come join us for a Meet and Greet at Norse Brewing in Wake Forest on August 12th at 6:00 PM! This is your chance to meet Paul, ask him any questions, learn about the campaign, and find out how you c...
000
Reposted by Matthew Mullins
Mark J. Nelson @mm-jj-nn.bsky.social · 05/08/2026
Cool: the Neurosymbolic AI Summer School has a logic programming tutorial. Cooler: The tutorial is by Bob Kowalski!
luma.com
Neuro-Symbolic AI Summer School 2026 · Luma
Please join us for the 5th Neuro-Symbolic AI Summer School (NSSS), a free, fully-remote meeting to be held Aug 6-7, 2026. Learn how to be part of this exciting…
131
Matthew Mullins @mmullins.coginiti.co · 05/08/2026
Kicking the Palantir habit seems like a good start. It's amazing to see someone complain about finding a good alternative. I can think of a half dozen off the top of my head that aren't actively trying to be evil. We don't ship spark in a box, but would consider it. 😂 #databs
politico.eu
Europe wants to kick its Palantir habit
The U.S.-based data crunching firm is deeply embedded in the most sensitive European workflows, from policing to intelligence to health care. Finding alternatives won’t be easy.
030
Reposted by Matthew Mullins
Andy Pavlo @andypavlo.bsky.social · 03/08/2026
👊I've joined the crew at @clickhouse.com to establish the *ClickHouse Labs* research group. We'll be working side-by-side with ClickHouse engineers to cook up ways to make the system even faster. I'll also be exploring new research topics related to PostgreSQL. clickhouse.com/blog/andy-pa...
clickhouse.com
Andy Pavlo joins ClickHouse to establish ClickHouse Labs | ClickHouse
I am excited to announce that I am joining ClickHouse, Inc. to establish and lead a new research team at ClickHouse Labs.
2676
Matthew Mullins @mmullins.coginiti.co · 28/07/2026
A million newly minted ontologists… and most of them have never even read Quine.
000
Matthew Mullins @mmullins.coginiti.co · 17/07/2026
Interesting opportunity, a weeklong workshop on ontological engineering with deep emphasis on BFO. Free in person and virtual summer school with NCOR. #databs ncor-network.org/summer-schoo...
ncor-network.org
Intensive Ontology Summer School | The Ontology Research & Development Network
PHI 605SEM (13336), The World's Ontology Ecosystem, August 3–7, 2026. Co-sponsored by the University at Buffalo and the National Center for Ontological Research.
010
Matthew Mullins @mmullins.coginiti.co · 17/07/2026
I don't watch much tv and even less reality tv, but this is surprisingly compelling. Watching 8 teams of engineers try to build a product is a good reminder that writing the code wasn't the hard part. @kelseyhightower.com is a fantastic host and it's hard to imagine a better choice!
1132
Matthew Mullins @mmullins.coginiti.co · 17/07/2026
Somebody make it make sense. Databricks is raising yet additional funding, beyond their last 16th round of funding, at a valuation that is north of 35x current revenue. #databs
databricks.com
Databricks is Raising a Strategic Round of Funding at a $188 Billion Valuation
The new capital will drive AI innovation across Unity AI Gateway, Genie, and LakebaseSAN FRANCISCO, CA — July 16, 2026 — Databricks, the Data and AI company, today announced strategic funding at a $18...
230
Matthew Mullins @mmullins.coginiti.co · 16/07/2026
In the last week I've had two calls with publicly traded tech companies where the CFO's hair is on fire regarding out of control token costs. I wish I could share company names, because even I was shocked by the spend level. We wrote up how a semantic layer can reduce token consumption for data.
030
Reposted by Matthew Mullins
coginiti.co @coginiti.co · 10/07/2026
the scary part is how good the wrong answer looks.
001
Reposted by Matthew Mullins
The A.V. Club @avclub.com · 08/07/2026
You can, and almost certainly should, opt out of being part of Muse Image's new experiment in making the internet a worse and more invasive place.
avclub.com
Instagram about to start letting internet randos "remix" your photos with AI
Instagram about to start letting internet randos "remix" your photos with AI
24334
Matthew Mullins @mmullins.coginiti.co · 01/07/2026
Two more reasons Coginiti 26.6's LLM blocks hold up: 1. Same block runs on Snowflake, Databricks, BigQuery, Redshift, Postgres, SQL Server, and with OpenAI, Claude, Gemini, or local Ollama. 2. Batching patterns keep prompts bounded, so it scales from 50 rows to 50k.
docs.coginiti.co
Process large datasets with LLM blocks | Coginiti Documentation
Learn four patterns for processing large datasets with LLM blocks while keeping prompts within token limits.
020
Matthew Mullins @mmullins.coginiti.co · 30/06/2026
The thing I like most about Coginiti 26.6's LLM blocks: the model's output is typed, so it's queryable immediately. JOIN it against real tables. Write tests that validate it. Cache it. Schedule it. AI output held to the same bar as every other transform in your pipeline.
docs.coginiti.co
Building an AI-Powered Data Pipeline with LLM Blocks | Coginiti Documentation
LLM blocks are scheduled to ship in the Coginiti 26.6 release. This tutorial is published as a preview so you can plan ahead — the feature is not yet available in earlier versions, and some details…
010
Reposted by Matthew Mullins
Christian Minich @christianminich.com · 29/06/2026
This by @mmullins.coginiti.co is so good. Summary: philosophers have had a place, especially around metaphysics, in tech, but with LLMs epistemology and ethics take center stage and are no long just a necessary foundation, but a product differentiator open.substack.com/pub/msmullin... #dataBS
open.substack.com
Philosophers in the Machine
Philosophers are not new to AI. From ontology engineering to alignment, philosophy has long shaped how machines classify, reason, and act.
093
Matthew Mullins @mmullins.coginiti.co · 29/06/2026
Coginiti 26.6 is out and you can now call an LLM directly inside CoginitiScript. New llm block type, sits next to your SQL. Write a prompt, declare a typed schema, get structured rows back, then query them downstream like any other table. The model's a pipeline step now, not a bolt-on.
061
Matthew Mullins @mmullins.coginiti.co · 27/06/2026
Beshear ‘28!!!
nytimes.com
Riding Into 2028: On the Midterms Trail, Andy Beshear Eyes a Bigger Prize
010
Reposted by Matthew Mullins
coginiti.co @coginiti.co · 26/06/2026
This is the conversation happening in a lot of AI analytics pilots right now. The demo works, the SQL is syntactically perfect, but then someone checks the number against the board deck only to discover the model invented its own churn definition and reported it with confidence.
031
Matthew Mullins @mmullins.coginiti.co · 25/06/2026
reminding myself to hate the game, not the players
060
Matthew Mullins @mmullins.coginiti.co · 25/06/2026
Her: It's hot outside today Me: Is the temperature 90 and rising? Her: Is that from a movie? Me: No it's the Partee Paradox, but we can resolve it with the Montague grammar. Her: But what about the Capulet grammar? 😂
000
Matthew Mullins @mmullins.coginiti.co · 25/06/2026
Listening to Tarkio, Colin Melody’s pre Decemberists indie-folk band, and I’d forgotten how good the songs were on this album. Here is a sample:
youtu.be
Tarkio - Keeping Me Awake (from Omnibus)
YouTube video by Kill Rock Stars
020
Reposted by Matthew Mullins
coginiti.co @coginiti.co · 24/06/2026
MIT CISR's new semantic layer briefing gets the problem right: AI value is gated by meaning, not data. ~1 in 5 orgs curate well; those that do are 3x likelier to get AI value. But it defines the layer as representation. The fix is operationalization. Why that gap matters 👇
coginiti.co
MIT Validates the Semantic Layer for AI Agents | Coginiti Blog
MIT CISR's semantic layer briefing gets the failure mode right: AI on ungoverned data fails plausibly, not loudly. But its own evidence points past representation to operationalization.
021
Matthew Mullins @mmullins.coginiti.co · 23/06/2026
I'm attending AWS Summit in Washington, DC. Let's connect and discuss semantic intelligence in the cloud.
030
Reposted by Matthew Mullins
Eric Schwitzgebel @eschwitz.bsky.social · 22/06/2026
Forthcoming soon: AI and Consciousness. Email me if you'd like to see the (uncorrected) proofs, or view the final public MS version at arxiv.org/abs/2510.09858
Cover: AI and Consciousness, by Eric Schwitzgebel
2296
Reposted by Matthew Mullins
coginiti.co @coginiti.co · 22/06/2026
"Semantic layer" means three different things right now: the one inside your BI tool, the universal one that serves every consumer, and the ontology from the semantic web. Same words with years of separate history. AI agents just made the difference expensive.
coginiti.co
Semantic Layer, Semantic Layer, or Semantic Layer? | Coginiti Blog
Three different technologies share one name. If you're evaluating a "semantic layer" right now, you need to know which one you're buying.
021
Reposted by Matthew Mullins
coginiti.co @coginiti.co · 16/06/2026
AI agents don't fail in production because the model isn't smart enough. They fail because they don't understand the data. Sharp panel with @bauplan.bsky.social on letting agents work safely on real data. Worth a watch 👇
youtu.be
Rethinking the Semantic Layer- Part II: The Builders Response
YouTube video by Bauplan
031
Matthew Mullins @mmullins.coginiti.co · 15/06/2026
Excited to be chatting semantic layers with fellow builders tomorrow at 12pm ET/ 9am PT. Bring your popcorn 🍿 because this promises to be a fun discussion. What even is a "semantic layer"? How do these semantic layers work? Do we still need semantic layers in the age of agents? #databs
luma.com
Rethinking the Semantic Layer- Part II: The Builders Response · Zoom · Luma
Semantic layers were built for a world where humans asked business questions, and the semantic layer helped translate those questions into a SQL query over a…
010