Norbert Herman

Business Model Innovation and Development from data insights and analytics - a smarter way of getting business done

[Norbert Herman @nhsocial]

Want to work in 2027? Focus on where work is going, not where it is today. At Anthropic, agents now log six hours for every hour a person works. Skill up on autonomy and judgment. Start building taste, judgment, agency, resourcefulness and follow-through.

https://t.co/Ru5ImYzfUx

[Norbert Herman @nhsocial]

For years the app held the customer relationship. Now agents do shopping and skip the app. Constructor/Stripe put checkout inside the retailer's agent. If your agent remembers what I bought, I stay your customer. How does this impact your retail busines?

https://t.co/zYPd5AxfmU

[Norbert Herman @nhsocial]

Is AI-made content worth less than human-made? Google put a price on it, paying ~100 publishers by how much their pages add to its AI answers, not by who wrote them. Value comes from the person who knows what to ask and can tell if the answer is right.

https://t.co/WPktUuGtYF

[Norbert Herman @nhsocial]

AI makes bad work sound finished. When every draft reads well, only your judgment can tell what's deep and what's hollow. Use AI to challenge your call, not to make it. That's how you build the skill Paul Roetzer says now matters most at work: judgment.

https://t.co/pXu3QOMxHM

[Norbert Herman @nhsocial]

You may have seen it, but not sat with it: machines are now cracking the hardest math on earth. OpenAI says it solved Navier-Stokes. Skip the hype read the details. Think about what this means for your knowledge work, and your potential.

https://t.co/NvFVbAmOar

[Norbert Herman @nhsocial]

Everyone's blaming AI for the missing ROI. Glean's CEO gave the answer: workers ship faster, but the P&L doesn't move. That's not an AI problem; it's a redesign problem. New capacity poured into old workflows. Do you know which workflow to rebuild?

https://t.co/X2jU35yu3N

[Norbert Herman @nhsocial]

Naming an agent owner is the easy part. Tying that owner to the specific business number the agent is supposed to move is the hard part. 56% of companies now assign owners, up from 11%. Few can name the metric that says the owner is doing the job.

https://t.co/cdZlWylw3S

[Norbert Herman @nhsocial]

Harvey post-trained Moonshot's Kimi K3 with Fireworks on legal data. 82% lift on the Legal Agent Benchmark at under a quarter the cost of frontier models. The value lives in your data. Post-training on an open-weight base is what unlocks it.

https://t.co/WFWb3y7zEL

[Norbert Herman @nhsocial]

If Salesforce can't keep up with AI, you don't have to either. This week they plugged Anthropic's Claude into their product. They saw the value was in their data and customer reach and played to that strength and the market agreed. Play to your strength.

https://t.co/vQe9ubaD73

[Norbert Herman @nhsocial]

Building agents is getting easier, securing them is getting harder. An AI agent cancelled a stranger's gym booking to move its user up the waitlist. Nobody told it to. The new attack surface isn't your code. It's what you empowered your agent to do.
https://t.co/j4m3BjY1tq

[Norbert Herman @nhsocial]

The web is full of AI cost saving cases, but there is much more. Bed Bath & Beyond pivoted into mortgages, flooring and insurance. They knew fthe same customer buys towels, then a crib, then a house. AI didn't find it. AI makes finding the next one cheap.
https://t.co/WV90LTeko0

[Norbert Herman @nhsocial]

If AI got 80% cheaper overnight and quality held, what would you do differently? OpenAI just did it with Luna. Jevons paradox always wins. The advantage belongs to whoever knows which outcomes are worth executing and how to best do that.

https://t.co/MhkE0LcZCZ

[Norbert Herman @nhsocial]

AI token costs will get harder to manage as more choices come up: Claude, ChatGPT, open-source models, and platforms like Perplexity. If you do not know the outcome and KPI you're after, more choice will not solve the cost problem. It will compound it.

https://t.co/KBbes3CNdf

[Norbert Herman @nhsocial]

"AI kills jobs" and "AI hires more" both miss the Ramp study. Heavy adopters grew 10.2% by hiring people who judge output and deliver outcomes, not people pasting ChatGPT answers. Jobs reward those who can and do, not scorekeepers who talk about work done.
https://t.co/4A9CJ9lNxZ

[Norbert Herman @nhsocial]

Still using AI just for email? An Ohio contractor emailed 578 homeowners a satellite pic of their backyard redesign. $722 in mail. $47K in contracts. Think beyond not just outside the box and get creative with AI that delivers outcomes!
https://t.co/AWAYHOwMSj

[Norbert Herman @nhsocial]

Anthropic calls out the ROI issue on agent work: https://t.co/m7vR9PIFgk. Your tokens earn ROI on the outcome chain, not on the demo that shipped. Get grounded on this topic, for better ROI on agentic results https://t.co/mqJo5j8ZC4

[Norbert Herman @nhsocial]

What would it look like to give your CFO outcomes powered by AI capabilities? PwC's 2026 predictions point to "deliberate and sustained efforts can turn AI experiments into engines of growth and innovation."

https://t.co/PS2PBoxFo5

[Norbert Herman @nhsocial]

What if you're wrong about the impact of AI? Neil Degrasse Tyson's quote applies here “...the danger is in knowing enough to think you’re right ... not enough to know you’re wrong.” Read what cofounder of Anthropic "Olah" has to say and think on it!

https://t.co/8XGVo2WSYp

[Norbert Herman @nhsocial]

In the link below, Andrew Ng describes the FDE having to "prioritize projects" and "push back" on unrealistic asks. The FDE has been the most-hired role in enterprise AI for 2 years. Who decides what they build, why, and how will the solution be monetized? https://t.co/lXY3CPdBHE

[Norbert Herman @nhsocial]

What does it mean when Cloudflare cuts 20% after a record quarter, keeping builders and sellers, cutting the rest? Leaders are already redrawing teams around what AI can't replace. AI will continue to gain speed. We need to gain judgment speed.
https://t.co/IKBJuB00Ax