CtxEngineer

Cloud Solution & Platform Architect | Electronic Music | believes in Radical Constructivism & in a secular state & decentralized identity

[Josh Cook | Microsoft MVP @FlowAltDelete]

I've noticed a lot of people still don't understand Microsoft Graph vs Work IQ.

Microsoft Graph is the data and action layer.
Work IQ is the agent intelligence layer.

> Graph: exact records, CRUD, integrations, background automation.
> Work IQ: cross-source context, synthesis, recommendations, agent workflows.

Work IQ sits ABOVE Graph. It does not replace it.

The winning pattern uses both:
- Work IQ interprets and plans.
- Graph executes and verifies.

Which layer are you building on?

[Microsoft AI Frontiers @ms_aifrontiers]

Today we're introducing Web Skill Factory: a pipeline that turns solved web tasks into reusable, verified code.

Most agent "skills" are notes a model rereads and reinterprets every run. Ours are programs.

On a WebArena subset with gpt-5.4, reusing the library moved accuracy on held-out instances from 55% to 70%, and cut average steps from 17.1 to 14.7.

[Brace @BraceSproul]

I just rewrote OpenWiki's entire prompt in v0.3 to generate more detailed & accurate wikis. Eval results:

- 📈 28.57% success increase (35% to 45% @ n=2)
- 🪙 14% fewer tokens
- 🧰 26% fewer tool calls
per successful task

And as you can clearly see from the graphs, the wikis contain much more data now.

Try out the new and improved OpenWiki: https://t.co/aZONneVM8A

npm install -g [email protected]

[Paul Iusztin @pauliusztin_]

The LLM isn't responsible for keeping your knowledge graph clean.

Your write pipeline is.

I split mine into two stages:

Resolution → normalize names
Deduplication → decide whether to merge

A wrong merge is much harder to recover from than a duplicate.

[Nicolas Camara @nickscamara_]

we built pdf-inspector so agents can process PDFs without waiting on OCR. it classifies any PDF in ~20ms and extracts clean markdown locally

→ 200 PDFs processed in 2.8s
→ top quality in extracting tables + graphs
→ built in rust
→ open source

https://t.co/Wjp9kpTHXJ

[Neo4j @neo4j]

Use #Neocarta to easily build semantic layer graphs for your agentic applications in #Neo4j

Alexander Gilmore shows how to build a semantic layer on #GCP using NeoСarta, a Neo4j Labs library that turns your data systems into a traversable graph.

How? Instead of hardcoding schema context into prompts, agents traverse from an entity to the right LookML view, confirm the glossary term, check lineage, and then issue the query. All in one traversal.

Worth reading if you are building agents on Google Cloud and hitting the limits of flat schema files: https://t.co/8pUaxxH2wJ

[Weaviate AI Database @weaviate_io]

Skip the transcript. Embed and retrieve the audio itself.

This notebook shows you how to do it with Weaviate and @GoogleDeepMind's Gemini in a few simple steps:

• Split raw audio into overlapping chunks
• Create multimodal embeddings with Gemini Embedding 2
• Store and search the audio in Weaviate
• Retrieve relevant clips using a text or audio query
• Generate answers grounded in the retrieved audio with Gemini 3 Flash

The example uses a recording of Robert Frost's "Birches," but the same approach can be applied to podcasts, interviews, lectures, call recordings, and other audio collections.

Explore the notebook:
https://t.co/q7WNOnbIHw

[The Year of the Graph @TheYotg]

Graph Databases Become Graph Engines

Building architecture before you buy a graph database and start adding all your data in it as Josef ‘Jeff’ Heusserer argues is a fair point. Plus, you may be able to use something like DuckDB or Lance DB or even Postgres or SQLite to address your graph needs.

But that doesn’t mean that graph databases are going away. What this shows is that database vendors are aiming to provide graph functionality, and get a piece of a $2.85 billion market, projected to grow $15.32 billion by 2032, exhibiting a CAGR of 27.1% as per Fortune Business Insights. As for graph databases, they are evolving and solidifying.

Case in point: benchmarking Postgres against Memgraph. Besides demonstrating how native graph databases differ when graph workloads get real, Memgraph also serves as a showcase for graph database evolution. 

Memgraph introduced Memgraph Zero, a product line built on the idea that data should stay where it is, and you should still be able to query it as a graph. In other words: a Graph Engine as part of a Graph Database offering.

“No pipelines. No ETL. No waiting for batch jobs before you can ask a question” is the promise, and the first product in that line, MemGQL, is available. MemGQL is a federated GQL query engine. It translates standard GQL into the native language of whatever backend, executes them in place, and returns unified results.

Neo4j announced its own graph engine product called Neo4j Virtual Graph, available in private preview. Virtual Graph lets you run Cypher queries and graph algorithms directly against the data you already have in Snowflake, Databricks, and other databases and lakehouses. Neo4j acquired GraphAware, expanding its footprint in intelligence analysis.

--  

📩 Excerpt from The Year of the Graph Summer 2026 newsletter  

Read "Layers of Meaning: Context Graphs, Graph Memory, and Ontologies for AI" with more sections, references and attribution here 👇  

https://t.co/Xq7SLcN7vd

All things #KnowledgeGraph, #GraphDB, Graph #Analytics / #DataScience / #AI and #SemTech.

[Neo4j @neo4j]

Wonderful news! Virtual Graph is now available in the Aura Console for Snowflake, Databricks, and Google BigQuery.

-Connect a source, generate a model, and run your first Cypher query in minutes. - - Public preview is free to use.

When we launched in private preview earlier this year, the signal from enterprises was consistent: they want graph intelligence on the data they already have, and they do not want to move it to get it. Virtual Graph is built exactly for that. Enjoy!

https://t.co/OAOaxawB9C

#VirtualGraph #AI #Cypher

[Kimi.ai @Kimi_Moonshot]

Releasing the model weights and technical report of Kimi K3.

Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window.

New model architecture: 2.5x the intelligence per unit of compute, not just more params.

Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale.

Model weights: https://t.co/7m7eEg6Y0B
Tech report: https://t.co/yeu6cjpMCT
Tech blog: https://t.co/YTfiMSNM1f

[CtxEngineer @ctxengineering]

@veitdengler Meine Parteimitgliedschaft endet dann nun ebenfalls. Meine Kernthemen wurden durch Strolz, Loacker und Sie hochgehalten - keiner mehr da. Schade…..

[CtxEngineer @ctxengineering]

@wikturna78092 @veitdengler Und wer wird die dann wählen? Das Wählerpotential ist dann bei 10-12% ausgeschöpft. Visionär, open-mindend spez. für Veränderung jegl. Art ist nur ein Bruchteil. Selbes in Konzernen - das Gros ist zu träge und hat Angst vor Verändeung; weils Lullis sind

[Bilgin Ibryam @bibryam]

NEW: Durable Functions in PostgreSQL 🤯

pg_durable is a PostgreSQL extension. The workflow state, queue, retries, and recovery live in Postgres, and it uses Postgres durability / HA / backups / recovery.

https://t.co/ommJ4lCUdH

[Femke Plantinga @femke_plantinga]

The traditional corporate wiki is officially dead 🪦

Teams are now shifting to agentic (self-updating) knowledge bases.

Simply put: data doesn't just sit there, but actively maintains itself.

So many companies waste months fine-tuning AI agents.

Then they wonder why the AI hallucinates out-of-date pricing.

The problem is not the LLM.
The problem is your stale documentation.

If humans can't find current info, your AI can't either.
And no, the solution is not a better "chatbot".

It’s a closed-loop context pipeline.

Here is The Self-Improving Knowledge Loop we’re looking at:
• Auto-Ingest: Pulling fresh updates from Slack, Jira, and GitHub, etc.
• Detect Drift: Finding exactly where documentation and reality split.
• Propose Fix: AI drafts the change, but never auto-publishes.
• Human Approves: The team retains absolute control over the source of truth.
• Ship + Re-Verify: The cycle resets automatically on a scheduled cadence.

This changes documentation from a heavy project to a 10-minute queue.

What is the biggest bottleneck in your team's internal knowledge right now? Is it ingestion or human review? 👀

Read more about this loop: https://t.co/yx8wbB36fX

[CtxEngineer @ctxengineering]

@Hippy15405085 @Thomas70327 So isses. Es wurde einfach nur sichtbarer.

[Azure Functions @AzureFunctions]

Building #Serverless #Agents on #AzureFunctions = writing a Markdown file.

YAML sets the trigger (timer, HTTP, Service Bus, email…), the body is the agent's instructions. No harness code - MAF handles it. Some examples:

@InfoQ's article covering it: https://t.co/gVIUfPlPaT

[Claude @claudeai]

New in Claude Code: Artifacts.

Interactive pages built from your session, like a PR walkthrough or a living project dashboard, shared with your team at a private link.

Available in beta on Team and Enterprise plans.

[Google Cloud Tech @GoogleCloudTech]

Introducing the Open Knowledge Format (OKF), an open specification that formalizes the LLM-wiki pattern into a portable, interoperable format.

AI is only as smart as the context we give it. As we build more advanced, agentic AI systems, they need accurate metadata and context to be useful. But in most organizations, that context is locked inside fragmented data catalogs, isolated wikis, scattered code comments, or the minds of senior engineers. Every time a new AI agent is built, teams are forced to solve the exact same context-assembly problem from scratch.

To solve this, we've announced OKF, a vendor-neutral, open specification that formalizes the "LLM-wiki pattern" into a portable, interoperable format. It provides a standardized way to represent the enterprise knowledge that modern AI systems rely on.

— Just markdown: readable in any editor, renderable on GitHub, indexable by any search tool
— Just files: shippable as a tarball, hostable in any git repo, mountable on any filesystem
— Just YAML frontmatter: for the small set of structured fields that need to be queryable: type, title, description, resource, tags, and timestamp

We’ve also shipped reference implementations to help you hit the ground running, including an enrichment agent for BigQuery, a static HTML visualizer, and live sample bundles on @github → https://t.co/ilhAMCrcTc

➕ Knowledge Catalog can now natively ingest OKF!

Stop reinventing data models and building bespoke integrations for every new AI tool. Here's more about how OKF works → https://t.co/FR4kJRsgEH

[James van den Berg MVP ☁ @JamesvandenBerg]

DoH is now generally available on #Microsoft Windows DNS Server 🚀
https://t.co/ess1Y1J9mL
#WindowsServer #Winserv #DNS #DoH #WindowsServer2025 #MVPBuzz #ITPro #AzOps #DevOps #SecOps @windowsserver

[Elli @ellishlomo]

Fable 5 is dead.

- Launched June 9.
- Killed June 12 by a US export order.

Anthropic pulled Fable 5 + Mythos 5 for everyone worldwide. Even their own foreign national employees are locked out, while other Claude models keep running.

This is a brutal warning shot. Imagine building your business on Fable 5 and watching it vanish overnight, with zero notice, because of a policy call you don’t control.

Anthropic says it’s a misunderstanding and they’re working to restore access. Maybe they do. But the point stands, it’s closed, and it proves how centralized AI can vanish at the stroke of a regulator’s pen.

If Fable 5 isn’t back by Monday, watch how the market starts repricing model risk, governance risk, and “AI as a single vendor dependency”.

This really is a fable with a moral: don’t bet your entire future on a black-box model you don’t own and can’t run yourself.