What GPT-5, SLMs, and new prompting frameworks reveal about the future of AI

🎭 Opening chaos: frozen screens and flying jokes

Sentiment: Playful, glitchy, high-energy
The meeting opened in classic office-hours fashion: Kyle froze mid-sentence, spawning a round of “is he a robot?” jokes. Greg compared GPT-5 to shifting from stick-shift to automatic. Kellye described it as “like a 5-year-old—literal to a fault.” Meanwhile, Ryan cracked jokes about robot potato-sack races, and the chat veered briefly into underground robot fight clubs and sushi-delivery bots.

Standout quote:

“Cooking an omelette is actually kinda tough.”

This was less about eggs and more about the group’s improv energy, where even breakfast becomes AI metaphor.


🧩 The GPT-5 question: built for nerds or everyone else?

A split-panel retro illustration. On the left, a smiling 1950s-style family gathers around a television labeled “AI,” treating it like a household appliance. On the right, hooded hackers hunch over glowing computer terminals, one screen showing a padlock icon. The contrast highlights mainstream casual use versus hacker-level intensity.Sentiment: Curious, skeptical, cautiously trusting
Lori relayed Conor Grennan’s take: GPT-5 isn’t for “the nerds like us,” but for mainstream users. Greg countered, saying he trusts GPT-5 more than any prior model. Erica raised worries about AI eroding critical thinking and reshaping testing.

Key tension: Is GPT-5 the geeks’ new power tool or a consumer-grade autopilot that drifts into the background of daily life?

Standout quote:

“No child born today will ever be smarter than AI.” – spotted in the wild


🔬 Compression shock: the 95% smaller LLM

into a smartphone. A shocked man holds the phone, while astonished 1950s-style onlookers—an elderly woman, a boy, and others—react with wide-eyed amazement. The background is textured like aged newsprint.Sentiment: Awe, future-shock, big implications
Jeno dropped a bomb: Multiverse Computing’s CompactifAI claims to shrink large language models by up to 95% without hurting performance. That means private LLMs running on phones, cheaper compute, and major energy savings.

Brendt expanded: the future may belong to SLMs (small large language models). Specialized, distributed, federated, SLMs could combine privacy with efficiency, breaking the cloud monopoly and putting AI in your pocket.

💡 Emergent insight: SLMs aren’t just about efficiency. They may rewire privacy laws, legal privilege, and even the energy debate around AI.


🔥 New frameworks: F.U.E.L. vs. CHEF

A vintage cartoon illustration on beige newsprint shows two chefs side by side. On the left, a smiling chef fuels a small car with a gas pump labeled “F.U.E.L.” On the right, another chef stirs a pot labeled “AI PROMPTS,” with steam rising into the word “CHEF.” The style is retro, halftone, and playful.Sentiment: Energized, practical, grateful
Kyle unveiled his new prompting framework F.U.E.L.:

  • Frame your intent

  • Use your voice

  • Edit & iterate

  • Launch thoughtfully

Half human discipline, half AI technique, designed to fight the “AI as vending machine” mindset.

Erica countered with her own acronym: CHEF (Context, Help, Example, Format). The overlap was uncanny, both frameworks cooking up a future where prompting is less theory, more applied craft.

Standout quote:

“Most AI, the first thing it gives you is going to suck. Don’t be kind to your AI if it’s bad, spit it in the garbage like Gordon Ramsay.” – Kyle

🧬 The architect emerges: Brendt’s pie-shaped knowledge worker

A retro halftone illustration shows a human silhouette emerging from the center of a segmented pie chart. The figure balances on one leg, juggling a pencil and a wrench, while a floating book and gear hover nearby. The background uses warm muted tones of orange, yellow, and teal with textured shading, symbolizing a multi-skilled “pie-shaped” worker.Sentiment: Visionary, disruptive, slightly unnerving
Brendt Petersen dropped the week’s deepest riff. His company is piloting AI-OS transformations in Brazil and the US, re-architecting orgs around architect roles: neurodivergent generalists with extreme pattern recognition.

Forget T-shaped skillsets. He’s pushing pie-shaped workers: breadth across multiple domains, with uneven but useful depth.

Brilliant quote:

“They’re the weirdos, the glue that held it together. But they trigger the corporate immune system when they rise too high.”

Brendt argues neurodivergence (autism, ADHD) may be an unfair advantage in the AI era. Architects take responsibility where others won’t, spotting problems before they explode.

🛡️ Security, lawyers, and local inference wars

a lawyer standing guard over a glowing AI locked in a briefcaseSentiment: Anxious, pragmatic, eye-opening
Privacy haunted the conversation. Ryan warned that lawyers, doctors, and engineers won’t let AI replace them because of licensing. Others flagged the legal paradox: if your lawyer uses AI to schedule calls, does attorney-client privilege still hold?

Brendt confirmed: with GPT-5 API access, contracts are shifting fast, especially since OpenAI became a military contractor. His advice: assume US-hosted models are monitored. For sensitive corporate work, pay the ~$3/day for self-hosted inference via Hugging Face, RunPod, or Colab.

Takeaway: AI adoption isn’t about features anymore. It’s about who controls the data, the contracts, and the context.

🎤 Closing sparks: gratitude & votes

As always, the room closed with energy and generosity. Erica dropped her Twin Cities Startup Week pitch link, Brendt fielded consulting side-questions, and Bob swore by his private Gemini instance.

Final vibe: A mash of improv comedy, legal drama, and visionary manifesto, where breakfast omelettes, robot fight leagues, and neurodivergence all pointed to the same truth: the AI future is messy, funny, and very, very human.

🔮 Summary

Vibe: Electric, glitchy, visionary
Biggest shift: Rise of SLMs + frameworks that humanize prompting
Standout moment: Brendt redefining the “architect” role for the AI workplace
Most quotable line: “Don’t be kind to your AI if it’s bad, spit it in the garbage like Gordon Ramsay.”

📦 Tools & resources from this week’s office hours

🛠️ Models & AI platforms
GPT-5 – latest OpenAI release
Gemini – Google’s enterprise AI
Claude – Anthropic’s legal doc helper

⚡ Breakthroughs & startups
CompactifAI – LLM compression (95% smaller)
Iterate.ai – AI innovation platform

🤖 Robots & real-world AI
RoboWar Detroit – underground robot fights
Cafe X at SFO – robot barista
Unitree Robotics – humanoid & quadruped robots

🔍 Hosting & inference
Hugging Face Models – 2M+ open-source models
RunPod – rentable GPU instances
Bolt Cloud – emerging platform for AI hosting
n8n – workflow automation

📊 Tracking & oversight
Tracking AI – monitor developments in AI