@Jakecusack

MCSE | MCSD | MCSA | MCPD Azure & M365 Architect | AI Engineer | Data Scientist

London, England
Joined July 2015
StudyScribe is live on the App Store today. Record any lecture, lesson, or voice note. On-device AI turns it into a summary, flashcards, and a quiz. Built with Apple Intelligence. Nothing leaves your phone. If we call it private, it has to actually be private.
1
102
Databricks' MemCurrent triples agent pass rate with memory
1
17
— Triple the pass rate, using fewer tokens, running faster: that's what happened when Databricks gave its production agents a proper memory architecture instead of no memory at all.
2
GitHub Copilot adds a model decision router
1
13
— GitHub Copilot now makes a decision before it makes a single tool call: which model should handle this turn.
10
Microsoft previews Copilot-curated SharePoint knowledge base
1
34
— Copilot is learning to compile a team's email, messages and meetings into markdown files its agents read before they act.
7
Postman narrows 170 tools to 15 for its AI agent
1
4
— Give an AI agent more than about forty tools to choose from, and it starts picking the wrong one.
2
Databricks gives coding agents their own database
1
8
— Two coding agents working the same feature will eventually collide over the same shared database, and Databricks just gave each one its own copy instead.
5
Microsoft brings CRM agents into Copilot and Teams
1
11
— Thirty new Copilot skills just moved customer-relationship work out of a separate CRM screen and straight into the chat window sellers already live in.
4
NVIDIA agents build Omniverse simulations
1
1
18
— One prompt just turned NASA's own public assets into a working 3D model of the International Space Station you can explore in a browser.
1
5
NVIDIA ships AIPerf, a rewrite of its LLM benchmark tool
1
13
— NVIDIA has published AIPerf, a ground-up rewrite of its GenAI-Perf benchmarking tool for LLM inference.
8
Databricks says the lakehouse now beats the warehouse
1
13
— Databricks says its lakehouse SQL performance is 77% faster than it was four years ago, and this time it's publishing the benchmark, not just the slogan.
9
Databricks runs its own reviews on Genie Ontology
1
16
— Ask a general AI agent a question about your own business, and it'll usually get it wrong, not because the model's weak, but because it doesn't know which tables are authoritative.