This week reminded me how fast the landscape keeps shifting.
One day it’s open-source challengers trying to outpace ChatGPT, the next it’s new training methods that cut human effort out entirely.
Tencent’s self-training approach and Sakana’s evolutionary algorithms both point to the same future: models that improve themselves without the costs we’ve come to expect. That’s a major shift in economics and speed.
Meanwhile, voice AI is having a renaissance.
After years of clunky assistants, the new focus is on expression and instruction-following. It may finally feel less like talking to a robot and more like collaborating with a capable partner.
The thread tying it all together is this: the cost of iteration keeps dropping.
That means our advantage as leaders comes not from buying the “best” system but from how quickly we can test, adapt, and scale what’s working.
The tools are accelerating; can you match the pace?
Let's dive in.
Patrick Patterson
CEO at Level Agency
What's interesting this week
Sakana AI’s Evolutionary Algorithm Breakthrough Models that evolve instead of retrain could blow up the economics of AI development. For growth leaders, this means more capable tools without waiting on expensive, slow updates. The question isn’t if this approach will spread, it’s how fast.
AI for Human Communication: Vocal Image Most AI talk is about efficiency, but this one’s about empathy. Tools like Vocal Image can help leaders, teams, and even entire organizations sound clearer and more confident. Imagine sales enablement with AI-powered coaching baked in. Read more
Hermes 4 Takes Aim at ChatGPT Nous Research dropped Hermes 4, claiming better performance without content restrictions. Open-source innovation like this keeps pressure on incumbents and reminds us the best model may not always be the most expensive. Read more
Tencent’s R-Zero: Self-Training LLMs Forget labeling armies. Tencent’s R-Zero shows LLMs can train themselves. This accelerates the race and makes data bottlenecks less relevant. Expect faster iteration and lower costs across the board. Read more
AI for Community Building Partnership on AI is asking the right question: can AI strengthen communities instead of just optimize ads? For leaders, the opportunity is to deploy AI in ways that build trust and connection, not just clicks. Read more
Deep dive: AI
Expressive voice AI is closer than you think
Voice AI has been hyped for years, but often dissapoints in practice.
The technology worked fine for basic commands, but the voices sounded robotic and the interactions felt stiff. OpenAI is now betting on two things that could change that: models that can follow complex instructions and speech that carries genuine expression.
The difference is more than cosmetic. A voice model that can take a nuanced request and respond in a natural tone is a different category of product. It’s the leap from “press 1 for billing” to conversations that feel alive.
That has obvious implications for call centers, sales interactions, and customer experience. The risk has always been that synthetic voices come across as fake or uncanny. If OpenAI clears that hurdle, adoption could accelerate quickly.
Customer listening has long been stuck in slow cycles.
A quarterly survey, a review of NPS scores, maybe a research project once or twice a year. By the time insights were gathered, they were already stale. Generative AI shifts that timeline from months to minutes.
By parsing massive volumes of customer conversations (support calls, emails, chat logs, and reviews), GenAI can surface objections, patterns, and opportunities in near real time. That means leaders don’t have to rely on hunches or lagging indicators. They can see what customers are actually saying and adjust strategy on the fly.
This isn’t about replacing human empathy. It’s about amplifying it at scale. The teams that build strong feedback loops now will outpace competitors still waiting on old-school reports. AI customer listening is becoming the new foundation of agile marketing and product development.
Nano Banana is Google’s codename for its Gemini 2.5 Flash Image model, now built into the Gemini app. On the surface, it looks like another AI image editor. In practice, it is a leap forward.
Nano Banana handles multi-turn edits where you can refine an image step by step while keeping the subject consistent. That solves one of the biggest pain points in AI creative work: faces and objects falling apart as you push them further.
I tested it myself and created a surprisingly convincing image of me standing on top of Mount Everest. No broken limbs, no uncanny face, just a clean, realistic result. It was fast and unsettling in how natural it looked.
Good Enough Isn't: Episode 2
The newest episode of the podcast is live, featuring Howard Diamond, our freshly minted CRO. This conversation is packed with lessons from his 16+ years in high-growth agencies and what he’s bringing to Level.
Howard and I dug into why strategy has to lead AI adoption, the cultural mantra of “good enough isn’t,” and why transparency always beats short-term gimmicks.
Level Agency is conducting a research study to understand how organizations are preparing for and adopting AI.
We’re collecting insights from teams across industries and we’d love your input. The survey takes 5 minutes and is designed to give your team a pulse on your AI readiness.
Once we’ve wrapped the study, we’ll share a full report with key findings and industry benchmarks to all respondents.
Special Bonuses:
Complimentary 90-day Action Plan: Organizations with at least 5 respondents will receive a free current-state assessment and a 90-day action plan to accelerate adoption.
Free 30-Minute Consultation: The first 10 responding organizations will receive a complimentary 30-minute consultation to discuss a use case and explore potential solutions.
