Alessandro Perilli @perilli

Original picture:

Original picture:

This weekend, I finally decided to build the very first Star Trek set ever produced by LEGO, which is, of course, the USS Enterprise NCC-1701-D from The Next Generation series (the only real Star Trek). The building techniques used in this set are mind-blowing: the designers had to go out of their way to recreate the many curves of the Enterprise-D with LEGO bricks. It really opens your mind to the possibilities still to be explored with this medium and makes you wonder how much more sophisticated LEGO sets will become as AI assists with their development. To show you the outcome of two people working for eight hours, I took some pictures, but there’s a twist. When I shoot on my iPhone, I usually use Project Indigo from Adobe Labs. When I opened the app this weekend, I discovered I’d been selected to try new AI-assisted editing capabilities. Sounded fun. So I went through the pictures I’d just taken, one by one, and asked it to apply a Cinematic effect. I expected a clever filter, similar to the ML Enhance feature in Pixelmator Pro. But no. Project Indigo changed the camera angle in each picture, completely relit the scene, and added lighting features to the LEGO set. The results are the two pictures you see above. To appreciate how far the AI went without hallucinating too many details, you have to see the original pictures, which I’ve posted in the comments. Very impressive in terms of both quality and speed. Now, time for Agentic Sunday.

keyboard_arrow_left Previous keyboard_arrow_right Next

If you are using agentic AI at the endpoint, the most transformative production boost comes from a feature I rarely see mentioned: annotations. It's even more useful than bidirectional live voice. That's how good it is. The ability to precisely tell my agent which element to change and how, in any document, web page, image, video, app in development, without resorting to multiple applications, has accelerated the way I work in a way I can’t overstate. I can't imagine going back.

@ZeroViKnight In the last six months, I often wondered what he would have said about the AI acceleration we are experiencing, especially about what happened during the Hugging Face incident. I wonder how he would have reconciled it with his late insight: competence without comprehension.

👀

Jeff Stibel @Stibel

5 days ago

Consciousness may have evolved to solve a problem that pure intelligence couldn’t handle, fascinating new theory suggests via @PsyPost psypost.org/consciousness-…

6 8 56 17K 87

event.idc.com/event/idc-ai-a…

My collaboration with Japan began almost 20 years ago, when Networld used to translate my work on virtualization[.]info into Japanese within 24 hours of publication. I continued collaborating with Japanese teams in other roles, and I’ve always been impressed by their strong work ethic, incredible attention to detail, and exquisite craftsmanship, even in software. So it’s with great pleasure that I return to Japan next month to speak at our IDC AI & Autonomy Forum in Tokyo. After the event, I look forward to meeting our clients there and the many CIOs who are transforming the Japanese economy with AI. 日本の皆さんとの協業が始まったのは、今から約20年前のことです。当時、Networldの皆さんは、私がvirtualization[.]infoに掲載した記事を、公開から24時間以内に日本語に翻訳してくださっていました。その後もさまざまな立場で日本のチームとの協業を続けてきましたが、仕事に真摯に向き合う姿勢、細部への徹底したこだわり、そしてソフトウェアにも感じられる見事な職人技に、いつも感銘を受けてきました。こうしたご縁もあり、来月、東京で開催する「IDC AI & Autonomy Forum」に登壇するため、再び日本を訪れることを大変うれしく思っています。 イベント後には、日本のお客様、そしてAIを通じて日本経済の変革を進める多くのCIOの皆様にお会いできることを楽しみにしています。

Yesterday, I got a chance to see the latest ByteDance models at the BytePlus Global Tour event in London. I left the world of synthetic media generation seven months ago to join IDC, so I was really curious to see how much progress had been made since then. After the event, a long, stimulating conversation about art, storytelling, and the economics of AI filmmaking with Henry Yan. @jiang4ai

It all depends on how accurate you have measured “completed task” and it's cost. Is it completed only in the sense that all steps have been executed? Or in the sense that the output from all completed steps is accepted by the company? Only the latter matters. And then: how do you measure that output quality? The cost of an evaluator for open weight models and for closed one is very different, and the time you reassess its performance changes, too. And this acceptable output we are talking about: that's not always flawless, right? So you have to factor in the cost of addressing the mistakes in the output that the evaluator didn't catch. And that is different, too, across open and close.

I’m hiring for two positions focused on AI Economics: one with an emphasis on financial modelling, the other on industry expertise at the business process level. For both, I’m looking for candidates who already use agentic AI in their current jobs or daily lives. These positions are not just about writing traditional analyst reports. We are building cool stuff at IDC ! I can promise unparalleled exposure to top decision-makers across industries and geographies. If you want to understand how the economy works and how it’s being reshaped by AI, reach out. DMs are open. (Colleagues and friends: if you know someone who could be a good fit, reshare this post or reach out to me in private. Ty!)

Curious about the impact of agentic AI on handicraft. Did anyone publish any data on how creators in the physical world are using AI for more ambitious projects? @Etsy, perhaps? I'd expect agentic AI to allow even people without experience or network to articulate product design, negotiate product specs, and set up supply chain, manufacturing, and distribution despite language barrier. And if that's true, I'd expect the most creative people to try and build more advanced products at a very small scale. Anyone?

Agentic AI for interior design. Not exactly the first use case that comes to mind when you think about agents and harnesses, but hugely valuable nonetheless. Over the last month, I’ve been using Codex to help me with several art installations. In each case, I had a number of elements that needed to be arranged permanently on a wall. Perfecting their distribution is as important as the quality of the individual objects themselves. It’s not simply a matter of calculating the coordinates of the anchor points across the available wall surface. You also have to find the pattern that best harmonizes the elements with one another and with the other elements in the room. This requires a great deal of experimentation and, once you find the right composition, a great many measurements. A model like GPT-6 Astra is impressive at representing objects with the correct dimensions, positioning them according to instructions, and preserving their relative sizes and proportions. Couple it with Codex, and it goes to the next level: It can recreate the entire scene in 3D using Blender. At that point, you are free to explore all sorts of alternatives without having to worry about recalculating the anchor points after every change. There are many more traditional ways to achieve similar results, but agentic AI puts advanced interior design simulations within reach of even the smallest projects. What frontier AI labs need to do next is let us skip the measuring part entirely. :)

@wintonARK Totally agree. Cost structure of AI initiatives should take into account both quality evaluation and hallucination/error remediation. Very few companies I talk to are doing that. Even fewer factor in the context window size into the business value realization equation.

The debate raging around AI today will forever be remembered as a textbook example of game theory at work.

I'm impressed by the capability to follow complex prompt instructions of the new OpenAI GPT‑Image‑2.5. But the model is still far from the creative and aesthetic sensibility imbued in Midjourney models. We'll have to see much bigger improvements to wow the Media & Entertainment industry. (This image required a lot of effort with GPT-6 Astra Mac and, eventually resorting to GPT-6 Astra Ultra to get the exact result I wanted. Yet, it's not perfect.)

Effective Agent Cost Management: A Practitioner's Framework for Planning, Sourcing, Implementation, and Operations: my.idc.com/getdoc.jsp?con…

Surprise! Even those end-user organizations that implement real-time cost monitoring for their agentic AI spend run over budget. In fact, almost 10% of them run over budget by more than 50%! Cost monitoring is not cost management. A dashboard gives you visibility into the problem, but solving it requires more. We recommended a cost management framework in our recent report: Effective Agent Cost Management: A Practitioner’s Framework for Planning, Sourcing, Implementation, and Operations. The link to the report is in the first comment.

The day humanity blinked.