We’re launching Go Agents - a workflow builder focused on maximizing agent accuracy on long, multi-step tasks.
We’re also releasing a generalist agent called Concierge, which delegates work to agents, acting as a central chief of staff for your users. Soon available via MCP.
The Jev Doom demo by @CompleteSkeptic is really cool, so while I’m waiting for access I tried the same thing with Gemma 4 E2B from @GoogleGemma.
Mine takes 640×480 frames directly, with no intermediate text description of the scene. It shares the image prefill across batched action queries, then reads the controls directly from logits.
It sort of works.
@gm_mertd@nathanbenaich SAP’s r&d budget is 7-8 bn, 1bn dedicated to AI can just mean renaming a few software r&d line items. Same language, different decade https://t.co/WCPWEDPXPu
AI generated slides often suck. @V7Labs created a slide generation framework that 1) makes beautiful layouts 2) deals with high information density, and 3) is fully compatible with Powerpoint/gSlides. Most V7 customers created tear sheets and memos with Gamma or GenSpark and got some pretty sloppy results, so we built this. Tools also matter - logo searching, generating background images, data plotting should all come out of the box. Some businesses create hundreds of slideshows a week, an we hope they'll produce them at scale with V7
News! @airstreet has raised $232,323,232 for Fund III to back AI-first companies from the earliest stages in the US and Europe.
Now the largest solo GP venture firm in Europe.
Our third epoch begins today. Join us!
I’ve thought about this a bunch, and don’t think its likely if we’re talking about silicon intelligence. Life is far more complex of a system so why invent something as complex as life + the entire periodic table of elements when you just need a few elements on a chip to solve the universe's purpose.
If the universe is a manufacturing process, its more likely to be a bio one. And if information is the goal, there is more information in the collective nervous system of pigeons than there is in all datacenters (which as we've proven with LLMs make a very approximate world model that fails to grok physics and simulate life). So if God had the goal of building superintelligence, it would be closer to a superorganism that can absorb higher entropy matter like connectomes and DNA rather than silicon, tin, and copper to make rectilinear chips which are just toys we're really proud of as evolved monkeys.
Most teams I talk to have the same problem: one person figured out how to use AI really well, and nobody else knows how.
The workflow lives in their head. It doesn't spread.
We just shipped something at @V7Labs to fix this, have a look.
We're adding more personality to the agent experience to curb loading times that can't be optimized further.
This month V7's Go agent got about 3x faster on time to first token, and a bit faster on tool calls. Now we're at diminishing returns, and it's better to invest in visual UX elements.
Wish I knew about it! Was working on my own mini game this morning. 3 independent agents per team need to collaborate to gather resources and build a city. This is GPT vs Gemini. GPT communicated more but one of its agent kept timing out (giving impossible next actions, basically tried cheating) and wasn't able to gather wood. Each turn they get a status of the map and any messaging from their teammates. One of those games that is more fun to build than to play it seems.
We're experimenting with workflow-powered apps. Our sales team asked if we could generate 3D floorplans before generating brochures, so I built a quick experiment last night to make high fidelity floor plan renderings from 2D images. Give it a try for the fun of it: https://t.co/pzNa8VA9wZ