"Semantic layers can't handle complex analytics." They aren't meant to replace SQL, modeling, or exploration. But complex work still needs stable definitions: which customers are active, what counts as revenue. Not a cage. The governed foundation you build on. coginiti.co/blog/semanti...

"Semantic layers are just metric stores." A metric without its business structure is incomplete. "Revenue" by what? Customer? Region? Recognized date? At what grain? Through which join path? Semantic layers model the structure, not just the formula. coginiti.co/blog/semanti...

"Semantic layers slow analysts down." Bad governance does. A good semantic layer speeds people up: no rediscovering the approved revenue calc, no reverse engineering joins, no asking five people which churn is correct. Restriction slows. Reuse accelerates. coginiti.co/blog/semanti...

"BI tools already have semantic layers." They do, and they were built to govern dashboards inside one tool. Governed meaning now has to serve SQL analysts, data products, APIs, and agents. BI semantics can't reach all of that. Enterprise semantics have to. coginiti.co/blog/semanti...

"The warehouse should own the semantic layer." Fine if you live inside one platform. Most enterprises run many, plus BI tools, notebooks, APIs, and agents. Trap semantics in one warehouse and consistency becomes lock-in. Enterprise meaning has to be portable. coginiti.co/blog/semanti...

"The data catalog is enough context." A catalog helps you discover and understand data. It does not define the approved measure, its grain, its join paths, or the executable logic to answer consistently. Discovery is not governance of execution. coginiti.co/blog/semanti...

The uncomfortable part of your AI bill: the price you pay is below the true cost of serving it. Frontier labs subsidize inference to win the market. Gartner's own analyst says they're losing money, and the savings won't flow to you. Budget for the meter.

"Agents can infer the business logic from the schema." Schemas encode physical structure, not meaning. A table named orders won't tell an agent whether cancelled orders count as revenue. Metadata helps it reason. A semantic layer tells it what you agreed things mean. coginiti.co/blog/semanti...

A Databricks engineer spent $7k on AI tokens in two weeks. Forbes calls it "tokenmaxxing." Gartner says token costs are on track to rival developer salaries. The token bill is an architecture problem, not a pricing one. coginiti.co/blog/semanti...

Token Costs Now Rival Developer Salaries. The Fix Is Architectural. | Coginiti Blog

AI token spend is climbing past developer salaries even as per-token prices fall. The durable fix is architectural: a semantic layer that cuts the tokens you spend at the source.

coginiti.co

A vendor-neutral standard convened by one vendor asks everyone to trust a venue they don't control. OSI is now Apache Ossie (incubating). The trademark, roadmap, and releases now belong to the community. coginiti.co/blog/apache-...

Meaning Belongs to the Commons: Why We're Excited About Apache Ossie | Coginiti Blog

The Open Semantic Interchange became Apache Ossie (incubating) in June. The spec didn't change—the governance did. For a standard whose job is to be trusted by competitors, governance is the product.

coginiti.co

the scary part is how good the wrong answer looks.

Documentation was fine when humans were the consumers, but agents don't ask clarifying questions. #databs

Coginiti 26.6 is here. We put AI inside the data pipeline. LLM Blocks in CoginitiScript make AI a first-class step alongside your SQL: generate synthetic data, classify, analyze sentiment, enrich text. Governed, testable, portable across platforms and models. Release notes 👇

Coginiti 26.6 Release Notes | Coginiti Blog

Coginiti 26.6 puts the model directly in the data transformation pipeline. LLM Blocks, a new first-class block type in CoginitiScript, let you embed AI reasoning alongside SQL in governed, publishable...

coginiti.co

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.

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.

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 👇

"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.

the AI agent, given full warehouse access, scrolling past your governed metric definitions to write a query against rev_amt_v2_final_OLD it's not bad at SQL. it's bad at knowing which of your six revenue columns is real. nothing in the schema tells it.

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 👇

Rethinking the Semantic Layer- Part II: The Builders Response

YouTube video by Bauplan

youtu.be

finance ARR ≠ sales ARR ≠ marketing ARR none of them are wrong. they're all "correct" against three different definitions. now ask an AI agent for revenue and watch it pick a fourth.

Anthropic ran an experiment most vendors would never publish. They gave their analytics agent direct access to thousands of prior SQL queries (every business question already answered correctly) and confirmed the agent actually read them. Then they measured the accuracy gain. 🧵

The webinar replay is up if you missed it. Christina Salmi (@ch1stna.bsky.social) and Matthew Mullins (@mmullins.coginiti.co) on the semantic layer imperative — 45 min + Q&A.

Closing the AI Readiness Gap, Part 2: The Semantic Layer Imperative

YouTube video by Analytics8

youtu.be

Let's talk about the myth of data consolidation. Many semantic layer vendors sell a story that quietly assumes your data lives in one warehouse. It doesn't. Real enterprises run 3, 5, 10+ platforms. Cloud + on-prem + lake + legacy. That's not going away.

Heading to Google Cloud Next in Las Vegas next week (Apr 22–24). If you're there and want to talk about governing business logic across BigQuery and multi-cloud environments — or just want to nerd out about semantic intelligence over coffee — reach out. Would love to connect in person.

Valerie Kwiatkowski just joined Coginiti as Director of Partnerships. Former CIO Americas at APM Terminals (Maersk), EVP IT at SDCCU, Managing Partner at StrataFusion, and Associate Partner at IBM Global Services. She's governed data at enterprise scale. Perfect fit.

"Define once, use everywhere" is a great slogan but a terrible implementation strategy. You push a metric out, and within 3 months someone says "this doesn't match our report." Fix it? Breaks downstream. Leave it? You're lying.

AI hallucinations in analytics are real: agents returning "confident" answers backed by undefined and unapproved logic. The fix isn't better AI. It's governed semantics. When your metrics are versiond, audited, and deployed with intention, your AI stops guessing. That's semantic intelligence.

We just joined the Open Semantic Interchange, an open source initiative to create a vendor-neutral standard for semantic metadata. Every tool defines metrics in its own format. OSI fixes that with one shared spec. This is what semantic intelligence needs to scale: open standards, not walled gardens.

Hot take: most enterprise AI projects will fail not because of bad models, but because of ungoverned business logic. Here's why.

We are proud to share that we have achieved three Google Cloud Ready designations for AlloyDB, CloudSQL, and Regulated & Sovereignty Solutions. Extending our ability to provide one door to data and help our customers run in compliance with programs such as FedRAMP or CJIS. bit.ly/CoginitiClou...