Why 93% of your AI budget is in the wrong place

Welcome back to The Cybernetic CxO, a monthly newsletter for CTOs, CIOs, and digital transformation leaders navigating the AI era.

93/7: The Budget Split That’s Killing Your AI Program

#

This number hit a nerve. My post about the 93/7 split (93% of AI budgets going to technology, 7% to people and process) reached over 10,000 people and sparked a real debate.

The pattern I see across organizations is painfully consistent: a team builds a brilliant demo, everyone celebrates, and then adoption stalls. Not because the model is weak. Because AI collides with:

• unclear decision rights
• messy workflows
• missing or messy data
• fear, skepticism, and zero time to learn

Here’s what the companies that actually scale AI do differently:

• invest in capability, not just software
• redesign the workflow, not just the UI
• define guardrails and operating model early
• train teams to collaborate with AI, not just “use a tool”
• measure success in outcomes, not tokens

AI is not a tech project. It’s an operating model upgrade. And your budget should reflect that.

I wrote about this in more depth in my new IT-Daily article: How Companies Become a Cybernetic Enterprise.

My 2026 AI Outlook: 8 Predictions

#

I published my AI Oracle 2026 in IT-Daily this month. The core thesis: the winners in 2026 won’t have the best model. They’ll have the best ability to deliver impact under real-world conditions.

The predictions in brief:

1. The AI bubble likely pops (macro-driven, not tech-driven)
2. Local models become the default: hybrid portfolios replace “one cloud, one model”
3. The Cybernetic Enterprise becomes the real strategy: feedback loops over silos
4. A Cybernetic Platform becomes non-negotiable: self-service, policy-as-code, observability
5. AI-native software engineering becomes the standard: controllability over speed
6. The next skill gap arrives: today’s hiring freeze is tomorrow’s capability gap
7. AI agents finally get productive: with engineering, not hype
8. AI shifts from productivity to relationship: emotional safety standards must come before rollout

2026 isn’t about having more AI. It’s about operating AI reliably, safely, and cost-effectively, at scale.

The Rise of the Agentic Enterprise

#

This month I sat down with Prof. Dr. Oliver Gassmann (HSG) for a conversation about what it really takes to move beyond AI hype into real business value. Watch the full conversation.

The key tension we explored:

• Agentic Enterprise = AI agents autonomously executing processes
• Cybernetic Enterprise = humans + AI steering the company through fast feedback loops

Most organizations are currently at what I call agentic islands: agents can execute 5 to 7 steps, then a human needs to refine, verify, decide. And that’s not a failure. It’s reality.

What is failing is the typical approach: “Keep the process. Add AI. Be surprised it didn’t move the needle.”

The path forward: map the value stream, simplify first, redesign end-to-end, and demand ownership. If an agent produces output, a human still owns the outcome.

OpenClaw: An Open-Source AI Agent Worth Watching

#

I spent three posts this month on OpenClaw (formerly Moltbot/Clawdbot), an open-source AI assistant from Vienna that’s become one of the most talked-about projects in the agent space (overview, architecture teardown, security hardening).

What makes it interesting:

• TypeScript CLI connecting to messaging platforms (Telegram, Slack, WhatsApp)
• Calls LLMs from Anthropic, OpenAI, or local APIs
• Executes tools and shell commands locally, in sandbox, or remote
• Memory system with hybrid search (vector + keyword)
• Agents can spawn sub-agents with isolated sessions

My hands-on experience: I wanted a “cozy Sunday project.” Instead, I spent an entire Sunday on security hardening. The installation worked fine, but a security review revealed several critical issues. Classic “works great, but…” defaults. Ports, mounts, root containers, credentials. All needed fixing.

Takeaway: Self-hosting AI assistants is getting easy. Operating them safely is still an engineering task. And that gap is where the real value lies.

AI Won’t Replace Developers. It Will Replace Your Software Process.

#

Most conversations about AI in engineering are stuck on speed. That’s the trap.

The real story is technical deflation: software is getting dramatically cheaper to produce. The winners won’t be the teams who type faster. They’ll be the teams who redesign the production system.

Dan Shapiro’s framing crystallized what I’ve been seeing:

• Level 0 to 2: AI as helper → intern → colleague (where most teams are)
• Level 3: Agent manager (your life becomes reviewing diffs)
• Level 4 to 5: Spec & orchestrate → dark factory

My take: the next competitive moat isn’t code quality. It’s spec quality + verification.

In a deflationary world:

• Specs become the new source code
• Tests become the new management layer
• Review becomes a product function, not an engineering chore

This is cybernetics applied to software delivery: tight loops, clear signals, automatic correction. So the system improves every run, not every quarter.

Would You Be Ready for a 60-Year Career?

#

My most-read post this month (11K impressions) wasn’t about AI at all. It was about longevity.

As lifespans increase, the classic model of education → career → retirement is becoming obsolete. We’re entering a reality of multi-stage lives and 50+ year careers.

Three trends that matter for leaders:

1. The Longevity Economy ($8T now, $12T by 2030) is reshaping markets
2. Multi-Generational Workplaces with up to six generations require age-inclusive leadership
3. Social Media & Longevity: 67% of Americans consider themselves biohackers

For organizations, this means rethinking pensions, phased retirement, reskilling programs, and workplaces as health hubs. Longevity isn’t just a health trend. It’s a leadership and business strategy.

My Current AI Stack

#

Claude Code: My primary tool now. CLI, coding, agentic work, writing, analysis. Everything. Works directly on my file system, no copy-pasting between tools. It fits how I actually work.

Perplexity: Web research with real sources.

NotebookLM: Feed it documents, get audio/video summaries.

Gemini: Image generation. Best and fastest.

Gamma: Slides and presentations. Fast, AI-native, and good enough to skip PowerPoint for most use cases.

ChatGPT: Canceled after being a user from day one. Output quality has been declining. I noticed the model often doesn’t even read provided documents or links unless you explicitly push it. Claude Code replaced it completely.

OpenClaw: Already abandoned. Security concerns were too significant, and it didn’t fit my workflow. A good reminder: “works” and “production-ready” are very different things.

Vibe Coded This Month (Private Projects)

#

Yes, I vibe code too. Here’s what happened this month:

• Website migration (continued): Still working on romanoroth.com and cyberneticenterprise.com. Found some very ugly things that need fixing. Vibe coding is so cool until you actually look at what it produced.
• My own AI assistant: Built it, explored it, already abandoned it. No time to maintain yet another side project.

Community Corner

#

DevOpsDays Zurich 2026: We received 387 talks, 47 workshops, and 28 ignite proposals. The program announcement is coming soon. Stay tuned.

DevOps Meetup Zurich: In February, we explored RTE exhaustion and on-call sustainability.

Sovereignty in Practice Study 2026: Together with VSHN, we launched a study on digital sovereignty in Switzerland. 5 to 7 minutes, open until March 31. Results will be shared at DevOps Meetup Zurich on June 3.

CAS Enterprise Architecture (HSLU): 3rd edition starts March 13. Now includes a dedicated AI application day. Info sessions: Feb 24, Mar 9, Jun 1.

CAS Digital Transformation (HSLU): Starts March 20. Learn to lead your organization through digital transformation. Info sessions: Mar 9, Jun 1.

CAS DevOps Leadership (HSLU): Starts April 24. Leadership and agile methods for modern software delivery. Info sessions: Feb 25, Mar 9, Apr 10, Jun 1.

Certified AI-Native Change Agent: Completed the training with Nikolaos Kaintantzis and David Baer. The biggest takeaway: turning AI hype into deliverable value requires aligning executives, engineers, and end users. And getting past the POC graveyard.