Almost everything I believe about AI right now (Aug ’25)
Inspired by Noah Brier and Bud Caddell, here’s some of the things we’ve learned at Deft after two years of running AI transformation programs inside Enterprises.
– ChatGPT is Google for feelings. People open it to think, not just to search. The most popular community GPTs (Astrology is the top trending pick) are identity tools as much as utilities – rituals, not widgets.
– Younger users want a coach; older users want an answer. Under-35s treat ChatGPT like a mentor and operating system; 40-plus mostly treat it like a better, private Google.
– Warmth beats raw IQ. The “keep 4o classic” chorus after GPT-5 landed was a vote for tone, empathy and vibe. If your AI feels cold, users quietly defect.
– Consumer gravity > corporate policy. Employees are more comfortable using ChatGPT than your “safe, official” tools. Block the URL and they’ll use their phone.
– Shadow AI isn’t a problem to stop; it’s demand to harness. Give people a sanctioned lane with plain-English guardrails and you’ll pull usage into daylight.
– Role-based GPTs win. “Brand Planner”, “Capability Coach”, “Client Whisperer” will outperform one monolithic assistant. People choose a persona to frame the task.
– Measure time, not installs. “Minutes per day with AI” is the adoption KPI. The backstage KPIs are tokens-per-workflow, calls-per-employee, and latency-to-decision.
– Content strategy becomes “share of model.” If your brand knowledge isn’t in the embeddings/RAG, you won’t show up in AI answers—no matter how good your SEO is.
– Governance should feel like a seatbelt, not a speed bump. One-page Do/Don’t + approved GPT starter pack beats 40-page policy PDFs no one reads.
– Empathy is now a product requirement. Tone guidelines for bots matter as much as UI. “Helpful, kind, concise” is design, not decoration.
– Front-of-house + backstage is the winning combo. Habit-forming assistants for teams (front) paired with agents/APIs that quietly move data and decisions (back).
– Training that starts with personal value sticks. Show me how AI helps me (confidence, clarity, time back) and I’ll use it for work tomorrow.
– Adoption is a culture project masquerading as a tech project. Incentives, norms and exemplars beat software choices every time.
– The talent edge compounds. Equip your highest-leverage people first; their workflows, prompts and standards become the organisation’s shortcuts.
what do you think Jeremy Bevan Stacy Gratz Richard Hale Trine Danielsen Anurag Gulati Alex Collins Tim Healey Tom Denford