As we press forward into the Age of Inference, at an exponential clip, it’s important we keep our ears to the ground and continue to listen for what’s on the edge and what’s emerging.
<aside> 💼
New Pricing Models for an automated world. How will SaaS pricing evolve in a world with AI agents handling tasks. Zendesk is pointing one clear direction - bill for when the agents work, not for the number of hours they’re on call.
New AI Business Model: Charging Customers Only When the Tech Works
</aside>
<aside> 💼
Where will the competition heat up? Here’s my Tl;DR for McKinsey’s recent scan of the places we’ll see the most activity, competition, disruption, new value creation…
McKinsey Report: 18 New Arenas for Competition, from Streaming to Space - InnoLead
</aside>
<aside> 🚥
TREND: Data Fabrics LLMs love unstructured data, but they can also be a source of errors. To ground AI, it’s helpful to structure data unambiguously, into a ‘knowledge graph’. One company doing that is WhyHow.ai. see their Case Study: “turning congressional transcripts into usable structured graphs”.
Case Study: Turning Congressional Hearing Transcripts into Temporal Knowledge Graphs
Also, beyond data lakes, now the growing trend of building an enterprise ‘data fabric’, organized yet unstructured, spread out across many places, but united by metadata, easy to compose and package for special applications in a dynamic manner.
Data fabric: Unlocking all of data’s superpowers
</aside>
<aside> 🚥
TREND: GenAI powered network monitoring
Agentic AI-driven automation within an intelligent operational framework, Dark NOC ensures proactive network management: End-to-end visibility, zero-touch resolution, resilient & self-sustaining network
The rise of the Dark NOC: a new era in network operations
Physical Network Inferencing (reasoning over electric grids)
How AI can prepare the power grid for the low-carbon era
</aside>
<aside> 🚥
TREND: XAI (explainable AI)
Building AI trust: The key role of explainability
</aside>
<aside> 🚥
SIGNAL: New LLM Architectures
<aside>
Thinking Beyond Models, into ‘Compound Systems’. Dr. Chris Potts from Berkeley AI Research Center lays out an important reframe for the future of AI development. Models alone don’t do anything, unless they are coupled to other parts. The whole system as a composition of elements is what to optimize for.
Stanford Webinar - Large Language Models Get the Hype, but Compound Systems Are the Future of AI
</aside>
NeSy, or neuro-symbolic, is an example of a ‘compound system’. The “neuro” part incorporates LLMs for part of the architecture, but fundamentally it’s processing concepts as symbolic constructs. Why? It requires much less data to train, and affords more accurate, explainable AI. https://research.ibm.com/topics/neuro-symbolic-ai
</aside>
<aside> 🚥
SIGNAL: GeoSpatial Inference (reasoning over terrains)
Applications of this technology will be seen in many settings, including urban planning, smart cities and crisis management. For example, imagine a hurricane sweeping across the Florida Peninsula and AI can pre-visualize the pre-disaster context, immediately survey post-disaster effects, identify resource and triage interventions, and with AI agents, immediately trigger and deploy solutions.
</aside>
<aside> 💡
Artist and software developer cum AI/ML explorer Sougwen Chung uses her own personal biometric data as input to a generative or algorithmic artistic process. She has curated her alpha brainwave states and visualized them with this robot.
https://substack.com/profile/710379-azeem-azhar/note/c-87816634Ar
</aside>
<aside> 💡
SIGNAL: Me+Machine
Longtime jazz musician and drummer JoJo Mayer spent his Covid downtime well. Besides practicing a lot, he hired a team of AI/ML developers to build a digital twin of himself that he could play and improvise against. He calls the combined unit Me+Machine and he is now actively touring and speaking about the project and its implications for all creative fields.
https://www.youtube.com/watch?v=URkxEAvavhw
"it listens to the input of my drumming and renders output according to how I play. It's like a backing track that listens to what I do. When I stop, it also stops. It renders variations on what I play and then I react to that. This creates a closed loop that uncouples the hierarchy of cause and effect"
"I think in this time we live in, where the future is coming at us with the accelerated speed and uncertainty, I think improvisation is possibly going to be a very helpful tool"
"the difference between human creativity and AI is that AI relates on interpolation, humans relate on extrapolation"
</aside>
<aside> 🧭 Consumer ‘Life Trends’ In 2025 This recent Accenture Life Songs look-ahead is excellent. Their work feels prescient, especially its emphasis on the growing need for people to re-balance digital and real-world time
https://www.accenture.com/us-en/insights/song/accenture-life-trends
My Tl/DR, https://www.innovationleader.com/report-tldr/accenture-lifetrends-key-trends-for-2025/
</aside>
<aside> 🧭 Mass Personalization (where genAI may shine) ”the customization of products and services at scale, using data and technology to cater to individual preferences while maintaining the efficiency of mass production.” Examples: https://www.arcade.ai/ “ask our AI models to design a thing for you” https://www.nike.com/w/nike-by-you-shoes-6ealhzy7ok
Bling It On_ An AI Startup Turns Shoppers Into Jewelry Designers — The Information.pdf
</aside>
<aside> 🧭 ‘Phygital’ (hybrid Digital/Physical Experiences) ”the seamless integration of physical and digital experiences to enhance customer engagement and create a unified, immersive brand interaction.” Example: 19 Crimes
Virtual ‘smart’ Mirrors https://digitaldm.com/virtual-mirror/ ****”interactive mirrors equipped with augmented reality and AI technology to provide personalized information, recommendations, and virtual try-ons for users” see: phygital showrooms
</aside>
<aside> 🧭 AI-Powered Procurement ”discover vendors to fulfill your unique manufacturing orders, generate RFPs, place orders, fulfill, all automatically, as your ai agent workforce works while you sleep” Examples: **Scoutbee, BabelusAI, **Booma.ai, Soource
</aside>