The Next Horizon: Automation Virtual Summit 2026
Why AI Projects Fail: Sol Rashidi on the Human Advantage
Session recap · Sol Rashidi
The Bobfather
Bob Schwartz is the CEO of Robert Mark Technologies and has led the company to consistent double-digit growth for the last 10 years. He is an active member of the Workload Automation community and holds a seat on the Broadcom Partner Advisory Committee.
At a Glance:
- Sol Rashidi said 74 to 88% of AI initiatives stop, pause or get canceled at the proof-of-concept stage.
- The fix: stop at licenses and start doing AI, which means redesigning the process, not just buying a chat tool.
- Sol’s advice on project plans: models and infrastructure at about 30 to 40% of the time, with the rest on people and governance.
- Data classification is the step you can’t skip before agents and automation scale.
- Sol’s message to automation teams: you matter more than ever, so use your judgment about what to automate.
Session recap · The Next Horizon: Automation Virtual Summit 2026
The Human Advantage: Trusted Orchestration for AI Amplification, Not Replacement
Sol Rashidi, World’s 1st Chief AI Officer, Best-selling Author
“Almost ready” might be the most overused phrase in enterprise AI. If your pilot has been almost ready for a while now, you’re in very good company.
Sol Rashidi gave the keynote at The Next Horizon: Automation Virtual Summit 2026, and it was all about why AI projects fail to leave the pilot stage, and why that makes the people who run automation more important than ever. Automation by Broadcom presented the summit with Google Cloud on September 24. Sol was there in 2011 when Watson took on Ken Jennings on Jeopardy, has served Fortune 100 companies as chief AI, data and analytics officer, and is now chief strategy officer at Cyera.
Stall at POC
74-88%
of AI initiatives stop, pause or cancel at proof of concept, per Sol Rashidi
Plan time on models
50-70%
Sol said it’s not uncommon to see this much project time on models and infrastructure
Sol’s suggested split
30-40%
of project time on models and infrastructure, with the rest on people and governance
Why AI projects fail: stuck in POC purgatory
Here’s the number that made me sit up. According to Sol, 74 to 88% of all AI initiatives stop, pause or get canceled at the proof-of-concept stage. Sol’s slide cited McKinsey (85% have not seen meaningful impact with generative AI), RAND (over 80% of AI projects are failing) and IDC (88% of AI pilots don’t reach production).
Sol calls it perpetual POC purgatory, and has been there. In an established enterprise with weaker data foundations and a ton of technical debt, AI isn’t as clean-cut as it is for startups and big tech. Feel like everyone else is moving faster? A lot more are stuck right there with you.
Four lessons learned the hard way
AI isn’t the answer to everything. Match the complexity of the solution to the business problem. To cut a piece of paper in half, why bring a chainsaw when scissors do the trick?
Using AI is not doing AI. Hundreds or thousands of Copilot, Gemini, Claude or OpenAI licenses used for emails and meeting notes won’t deliver the productivity lift, Sol said, and reports from McKinsey, PwC and MIT show organizations aren’t getting the return they were promised. “The real ROI comes from doing AI.” That means breaking a function like FP&A, procurement or supply chain into tasks and subtasks, redesigning the process, then deciding where automation, agents, generative AI or applied AI fit.
Flip the project plan. Sol said it’s not uncommon to see 50 to 70% of project time spent on model selection, infrastructure, RAG, fine-tuning and vector databases. Sol would put models and infrastructure at about 30 to 40% and spend the rest on workforce preparation, AI fluency and literacy. (Three hours of Copilot training doesn’t count.) Those are Sol’s forgotten pillars of success, along with AI and data governance and security.
Not all AI is created equal. Governance for applied AI is fundamentally different from governance for agentic AI, Sol said, and lumping them together is a common mistake.
Agents and automation are the connective tissue
We still treat people, apps, databases, models and agents as loosely connected pieces, Sol said, but it’s becoming one big mesh, and the connective tissue is agents and automation. Not feeling it yet? Sol expects you will in the next six to nine months.
That’s why Sol sees automation getting a resurgence, the way data does every two to three years. Your role is growing with it: you own strategy as well as execution, including what to automate, how, and what data and context to bring in.
Sprawl, AI governance and the step you can’t skip
Sol’s running joke: ask six people to define AI governance and you’ll probably get 16 different answers. Right now there’s no playbook and no one-size-fits-all tech stack. It’s like The Cheesecake Factory menu, Sol said: everyone orders something different, and all the answers are right.
Meanwhile, the business wants speed, so instead of one coordinated automation platform, we get dozens of orchestration tools, scripts and monitoring systems, with agents and workflows running independently across teams. The fallout, Sol said: inconsistent governance and security, more risk, service levels not always met, frequent failures and interventions, and murky accountability.
The step Sol says you can’t skip is data classification. Agents and automation depend on access, data and context, so know what’s sensitive, confidential, internal-only or OK to share externally. Then you can trace data from source to target, see who has access and how it’s used, and switch things on or off. Governance has grown, too: it now covers data in motion (network traffic, APIs, file transfers) and data in use (prompts and agentic workflows), not just data at rest.
The thankless heroes
Sol’s closing was for you. Boards and executives ask why it isn’t done already, without seeing the data, governance and security work that makes an agent or workflow reliable at scale. Sol called the people doing it thankless heroes. Sol’s advice:
- Use judgment about where to automate, and where to hold off because the data foundation isn’t stable enough.
- Pick partners with security and governance in their DNA. This is about reliability and stability, not cool features.
- Let AI handle the execution while you handle the thought. My favorite line of the keynote: “Outsource tasks, not your critical thinking.”
- Be human-led and AI-supercharged, not the other way around.
One parting thought: technology can scale efficiency, but it’s relationships that scale opportunity.
What this means for you
The unglamorous work is the important work. Redesign the process instead of stopping at licenses, classify your data before you scale, and use your judgment before you automate. Next, read our recap of Serge Lucio’s session on agentic AI without the chaos and Caroline Denoon-Slater’s blueprint for AI agent governance.
So here’s the question to take back to your team: how many tools, scripts and schedulers quietly run your business today, and does each one have a clear owner?
RMT has spent 25 years in workload automation, with 1,500+ engagements behind us, and we’re glad to be a second set of eyes. Book a free 30-minute assessment or explore Broadcom’s workload automation solutions with us.
Not sure what to automate first?
RMT helps automation teams sort out what’s running, what’s worth automating and what governance to set up before AI gets involved.
More from the summit
The Next Horizon: Automation Virtual Summit 2026, session by session
- Why AI Projects Fail: Sol Rashidi on the Human Advantage (you’re here)
- MCP Server Security: Agentic AI Without the Chaos
- Intelligent Control Plane in Practice: 4 Automic Automation Use Cases
- Evertec’s VM Rightsizing Automation: From Manual Tickets to One Automic Workflow
- The Orchestration Accountability Gap: Inside the SLA Illusion
- From Black Box to Glass Box: A Blueprint for AI Agent Governance
- The Illusion of Green: Automation Observability Beyond the Return Code
- Broadcom Automation Roadmap: What Shipped and What’s Next