5 Key Questions Every AI Leader Needs to Ask Themselves

The AI platform wars have escalated into a full-blown arms race.

• OpenAI’s ChatGPT 5.1 brought dynamic thinking time and tone-based personalization.
• Anthropic’s Claude Opus 4.5 arrived optimized for multi-hour, multi-agent projects.
• Google’s Gemini 3 Pro is not just leading the LLM leaderboard but delivering a one-million-token context window and deep Workspace integration.

But even as models clear new performance ceilings every month, organizational productivity is flattening.

The McKinsey ​analysis​ we ​highlighted recently​ shows that today’s AI can already automate 57% of U.S. work hours—a potential $2.9 trillion value that could be unlocked by 2030.

Yet, fewer than 40 percent of companies report measurable gains. And the constraint isn’t the tech itself, but us and our organizations.

So how do you ensure your organization ends up on the right side of the emerging productivity divide?

Let’s walk through the five questions every AI leader should ask themselves in a bit more detail.

The Five Questions

1. Are we effectively redesigning our workflows to unlock the potential of AI, or are we simply overlaying new tools onto old processes and risking the "productivity slowdown?

Every transformation either accelerates or slows down, depending on where you start.

​Stanford’s recent comparison​ of human versus AI-agent workflows captured the danger well: full automation attempts slow human teams down by 17.7% due to errors, hallucinated data, and misused tools that require extensive human verification.

On the other hand, targeted augmentation, like delegating specific steps, accelerates work by 24.3%.

But to get there, we need ‘workflow literacy,’ we need to understand how work actually happens.

The McKinsey study mentioned above echoes the same pattern.

When organizations focus on tools rather than workflows, impact stalls, but when they redesign around people–agent–robot partnerships, value accelerates.

They analyzed 7,000 skills and found that non-physical knowledge work, dominant in modern enterprises, is overwhelmingly automatable by agents.

But the bottleneck is the redesign itself.

As Gartner warns in a ​new framework​, staying on the left side (simple automation or augmentation) will deliver incremental gains. The right side (transformation) is where long-term advantage emerges.

Figuring out which company you want to be, and then designing your change processes toward it, is an exercise every leadership team should undertake. (I’ll be on the right side with our company.)

2. Are we building a culture of learning that starts at the Board level, or is AI fluency treated as a mandate only for the workforce?

Culture comes from the top, and ServiceNow provides the clearest ​case study​ of the moment.

CHRO Jacqui Canney described their approach as a “capability, confidence, and leadership shift,” and operationalized this by having the Board take the same AI capability assessment as all 28,000 employees.

The broader environment is shifting quickly as well: demand for AI fluency has grown roughly seven-fold in two years and now appears across seven million U.S. jobs.

Despite this surge, many leadership teams still assume AI literacy is something “the staff” must figure out.

The companies pulling ahead (and we love working with) are flipping the script: learning starts at the top.

3. Are we equipping our managers to lead hybrid teams of "people, agents, (and robots)," or are they still managing based on human-only metrics and expectations?

Managerial capability is the least discussed yet most impactful way to drive AI transformation.

We are entering a world where managers will orchestrate hybrid teams whose members include full-time employees, part-time contractors, model-driven agents, and soon, physically embodied robots.

Yet most managers still operate with human-only metrics and expectations.

That mismatch will widen quickly.

Google’s Gemini 3 Pro ecosystem, which now includes ultra-high-accuracy image generation via Nano Banana Pro, research-grade automation through NotebookLM.

Especially end-to-end workflows made possible via Workspace Flows, show what "agentic teammates" will soon contribute: deep research, multi-step reasoning, live data retrieval, and autonomously generated outputs (like the real-time city-weather visual or Eric Pores’s career storyboard below).

As Claude Opus 4.5 demonstrates with its multi-hour task leadership and new computer-use capabilities, these agents are no longer “assistants.” They are colleagues operating programmatically and at scale. Managers need new playbooks—how to scope work, verify outputs, benchmark agent performance, and redesign roles in real time.

4. Is our proprietary data actually accessible to AI tools, or is it locked in silos and "people's heads" where AI cannot reason over it?

A ​new report from Superintelligent​ put numbers to a problem every leader feels: nearly 45 percent of companies lack documented workflows, and even more suffer from severe data fragmentation.

And, this is crucial: agents cannot automate what only exists in our memory.

This is why Superintelligence frames 2026 as “the year of context”: clean data, unified access, and documented workflows as preconditions for agentic automation.

It’s also why Google’s Workspace-native ecosystem (as does Microsoft with Copilot) poses such a strategic threat to third-party platforms like ChatGPT.

When your data already lives inside Workspace, Gemini 3 Pro can reason over it natively, no integrations required.

If your data is locked in PDFs, scattered across legacy systems, or stored in the heads of veteran employees, you can’t be an AI-ready company.

5. Do we have a clear mechanism to encourage development and identify local experiments and scale them across the enterprise?

Most enterprises still treat AI as a glorified Q&A interface.

But the frontier has moved. Claude Opus 4.5 can now lead multi-agent, multi-hour projects. NotebookLM generates full research dossiers with citations, auto-builds slide decks, and synthesizes PDFs, Sheets, images, and URLs.

As mentioned above, Workspace Flows allows no-code automation linking Gmail, Docs, Sheets, and Meet into live agentic workflows, like my pre-meeting research agent that autonomously prepares a deep-dive client brief 10 minutes before every call.

The question is no longer whether (agentic) AI workflows work, but whether you have a mechanism to identify promising local experiments and scale them across the enterprise.

In the ServiceNow case study, we saw how they crowdsourced 1,000+ use cases, then narrowed them to 27 high-value projects.

You need to build an AI project pipeline like this, too.

The research we shared should show that the gap between winners and laggards will not be determined by who has access to the best model but by who can redesign work, culture, and data flows fast enough to capitalize on them.

If you’re answering “no” to any of these, reply, and I’ll share ways to quickly make progress toward a yes.

Did the research trigger even more questions? Get in touch as well – we love to keep working at the speed of AI.

Until next week,

‍

- ​Daan