Databricks Genie is now available inside Slack, letting teams ask questions about their business data and get answers with charts and files.
I like this because useful information shouldn't require knowing exactly where a dashboard lives or which report to open.
Being able to ask a question in the place your team already works could make data much more accessible. The important bit is making sure the answers are grounded in the right data
1
19
Adobe bringing its Creative Agent into Premiere and InDesign is an interesting step for creative workflows.
The useful part is the ability to handle multi-step edits instead of making you jump between tools for every little adjustment. Imagine spending less time on repetitive production work and more time shaping the final result.
I'm still a believer in keeping the human in charge of the creative decisions. Let AI handle some of the busywork, not the taste
1
23
Microsoft's global outage is a reminder that AI productivity depends on more than having a capable model.
Outlook, Teams and Copilot were disrupted across multiple countries, leaving businesses struggling to manage emails, meetings and everyday work.
It's easy to get excited about AI agents taking on more tasks, but reliability matters just as much as intelligence. If the tools and infrastructure behind your workflow go down, all that productivity disappears with them.
I love seeing AI become more useful at work. I also want to see companies build proper backup plans around it. We can't automate everything and forget about resilience
1
1
140
Google just introduced Gemini Decision 1, a model designed to make fast yes/no, multiple-choice and scoring decisions instead of generating paragraphs of text.
I like this approach because not every AI task needs a long conversation. Sorting support requests, checking documents against a rubric or routing work to the right team are often decision problems.
Using a smaller, specialised step for those jobs could make AI workflows faster and cheaper. Sometimes the best assistant is the one that knows when not to write an essay
1
41
One thing worth remembering about AI productivity: using more AI doesn't automatically mean getting more done.
Gallup's latest workplace research points to better results when AI fits into existing workflows and managers help people use it effectively.
My takeaway: don't collect AI tools just because they're new. Find one repetitive task, build a workflow around it and see whether it gives you time back. That's a much better measure of productivity
1
2
30
Claude can now join Google Meet calls, follow the discussion and help turn ideas into plans and timelines.
I like this direction for AI at work. Meeting summaries are useful, but having an assistant help organise the next steps while the conversation is still happening could save teams a lot of back-and-forth.
I'd still want people checking the action items before they go out. Nobody needs an AI confidently assigning the wrong task to the wrong person 😅
2
33
this is a smart direction for AI. not every task needs a long generated response, sometimes you just need a reliable decision at scale. the pricing is interesting too, especially for teams processing large volumes of data
Today we’re launching the Gemini Decisions API, powered by our new Gemini Decision 1 model in @GoogleAIStudio.
Instead of generating text token by token, it’s built for closed-set choices scoring up to 1,000 questions across text, images, audio, video + PDFs in a single pass. A few details:
- 3 decision types: yes/no, multiple choice (up to 255 options) + rubric scoring
- $0.10 / 1M input tokens with $0 billed for output tokens (~5x cheaper than Flash-Lite)
- leads on several benchmark evals with lowest calibration error
Google teams are already building with it, below is a look at Stitch Rapid designing a full product flow in seconds (1x speed)
1
16
OpenAI is adding verified business profiles and AI agents for US e-commerce companies inside ChatGPT.
This is an interesting shift for small businesses. People may increasingly discover a product, ask questions and compare options without ever opening a traditional search page.
That also means businesses will need to pay attention to how accurately their information is represented by AI. Being discoverable is one thing; giving customers reliable answers is another
2
2
40
IBM and Microsoft are teaming up on Imagine Enterprise, an AI agent platform aimed at regulated industries.
The interesting part is the focus on putting agents into existing business systems, with consulting support and Microsoft's cloud and security stack behind them.
For enterprise AI, the hard part isn't getting an agent to complete one impressive demo. It's making it dependable across real workflows, permissions and compliance requirements
1
26
a munich startup called marble cognition just raised $3.75m on a pitch every normal person can learn from: most companies' processes aren't well enough defined for an ai agent to take over. so instead of handing agents a vague job, marble watches how a team really works first, then builds agents around what it saw.
that tracks with my own experience. a common reason an ai tool disappoints someone isn't the tool, it's that the task lived in their head. my weekly reporting "process" used to be fourteen browser tabs and vibes. no agent could have taken that over, and neither could a new hire.
here's a 15 minute test you can do before you automate anything. write the task down like you're training someone on their first day: what triggers it, what you need before you start, the steps in order, how you know it's done, and what to do when something goes wrong. anywhere you catch yourself writing "it depends" or "i just know," that's the part an ai tool will get wrong. fix that gap first, or keep that step for yourself.
same week, the guardian ran a headline about 29 nhs trusts said to be benefiting from palantir software using it little or not at all. buying a tool and changing how work gets done are very different projects. (i'm going by the headline only, so treat that as a pointer, not a verdict.)
so before you shop for another ai app this weekend, spend 15 minutes writing your process down. it's the least exciting productivity tip i have, and it's the one that works
4
2
70
OpenAI's acquisition of Pierre is worth watching from a productivity angle.
Pierre's tools help developers navigate files, inspect changes and review code. those are the parts of coding work that become even more important when agents are generating changes in parallel.
good AI workflows need good review workflows too
33
Anthropic's AI development pace may have doubled over the past year, according to independent researchers at METR.
one explanation they point to is AI helping build newer AI systems. that's fascinating, but it also makes independent evaluation and safety testing more important as capabilities improve
16
this could make personal finance feel a lot less intimidating. asking questions about your own spending in plain language is useful, and keeping access read-only is a smart boundary for a tool handling sensitive financial data
Introducing Personal Finance in Claude, in beta in the US.
Connect the accounts you choose, and Claude answers your money questions in everyday language, from your own balances and transactions. Access is read-only, and Claude can't move, manage, or invest your money.
11
the proactive part is what stands out to me. having ai teammates that can take ownership of tasks, use your existing tools and collaborate in slack could make a real difference to how teams get work done
Introducing teammates.
Teammates work for you. They take on jobs, and work proactively using your Computer tasks, skills, memory, and tools. Teammates collaborate with other teammates and can message coworkers in Slack.
Available now for Max subscribers on the web.
5
211
Sam Fender covering the Rolling Stones for the John Lewis Christmas advert is a pairing I didn't know I needed.
the choir from his old school and the brass band give the recording a personal touch. a familiar song can feel completely different when the arrangement has a story behind it.
1
212
a Stanford study reported that many widely cited claims about LLM capabilities weaken or become outdated surprisingly quickly.
it's a good reminder to treat benchmarks as snapshots, not permanent labels. models move fast, and yesterday's leaderboard doesn't always tell you what a tool can do today
iyooo.com/en/news/day/2026-1…
19
Hugging Face has introduced Model Integration Contracts to automate testing for Transformers integrations.
i like this kind of developer tooling because it helps catch compatibility problems earlier. when teams are working with lots of models and libraries, reliable testing saves time that would otherwise go into chasing unexpected breakages.
1
18
Amazon Bedrock Knowledge Bases is adding Crawl4AI integration, giving developers another way to bring web content into their AI-powered knowledge systems.
this is useful because a lot of AI workflows depend on getting information into the right format before a model can do anything with it. better collection and preparation can make the whole process smoother.
19
turning pitch decks and screenshots into a polished launch video could save teams a lot of production time. curious to see how much creative control it gives you.
Introducing Launch Videos in OpenArt!
Turn your website, screenshots, pitch decks, and brand assets into a stunning product launch video, complete with cinematic transitions, motion graphics, animated text, and music.
Try it now. Link is in the replies!
14
🫠
LangSmith Bring Your Own Cloud (BYOC) now runs on @AWScloud, @Azure, + @GoogleCloud.
Your traces and agent deployments stay in your cloud account while we run the data plane.
☁️ GA on AWS, Beta on Azure + Google Cloud
🚚 Moving from self-hosted? We'll run the migration
langchain.com/blog/langsmith…
8




















