Not your engineers.
Not the models.
Not the tooling.
Your executives.
If your leadership team still runs the company through slide decks, your AI initiative is already compromised.
AI does not thrive in PowerPoint.
Work becomes about presentability:
- what fits the slide template
- what tells a clean story
- what avoids ambiguity
One Company, Two Realities
Every company has two operating systems:
- doing the work
- reporting on the work.
Finance works inside Xero models and reconciliations.
Sales lives in Salesforce pipelines and probabilities.
Engineering lives in GitHub repos and Jira boards.
These environments are dense. They contain trade-offs, dependencies, edge cases, and risk signals. They preserve state. They evolve daily.
Then there is the second operating system: Slides.
Before anything reaches the executive layer, it must be digested to pass through a membrane. Context is compressed. Complexity is flattened. Risk is color-coded. Narrative replaces state.
At first, this feels harmless. Summaries are useful, but the summary becomes the schema. The medium is the message.
The PowerPoint Gravity Well
You summarize.
You abstract.
You simplify.
Your processes begin reorganizing themselves around reporting. Ask any support leader why tickets are categorized a certain way. Ask any product team why agile roadmaps are structured into quarters. Ask finance why KPIs are grouped into those specific buckets.
You will hear the same answer:
“That’s how the board wants to see it.”
The slide are the schema. Red-yellow-green is more real than the underlying situation. KPIs are reality.
Seasoned executives know about Mcmanara Fallacy and Goodhart’s Law. But, a RAG in a PowerPoint causes them to forget both.
The US military famously banned powerpoint in 2010, because of the impermeable membrane between it and reality.
When metrics become the presentation, they become the reality.
Processes get shaped for narrative clarity, not execution. Data gets organized for dashboards, not intelligence. Decisions drift toward permission-seeking instead of system steering.
This was survivable… before AI.
Dreaming in No-Code (Again)
With LLM agents, everyone predicts a revolution in engineering. There will be one, but not the one executives think.
Engineers have always been the bottleneck in software development. We know this. Engineers do one thing: we automate things. The thing we love to automate the most are the boring and repetitive tasks we do ourselves. For decades we have been building tools to make developing easier.
Some of these tools were aimed at enabling everyone to be a developer.
COBOL. In 1960, COBOL promised that anyone could talk to computers in simple syntax. Programming was democratized. Except it wasn’t. Typing wasn’t the hard part. System design was.
Low-Code/No-Code Platforms: The UI would replace expertise, thousands of platforms promise full apps in minutes. It worked, for predefined problems. The moment you stepped outside the template, you needed abstraction, constraints, and modeling again.
The System Bottleneck
Coding has always been a bottleneck, but not because typing was hard. Easier syntax and automated UI’s didn’t solve the actual problem.
We don’t have a coding shortage. We have a system-design shortage, the rare and tedious ability to translate messy, ambiguous reality into structured systems:
- Generalize patterns
- Define constraints
- Specify edge cases
- Decompose complexity
- Decide what matters
We don’t lack people who can ask ChatGPT to write code. We lack people who think in systems.
The Irony of LLMs
The irony is that the most powerful coding tool in history doesn’t point and click. It reads and writes.
AI was created for the command line.
LLMs were born in developer ecosystems. We software engineers have actually had a lot of success in automating engineering tasks. As a field, our productivity has kept up with Moore’s law. In the 1960’s, a developer might manage several thousand lines of code; now we work with millions!
Serious teams are not “prompt hacking.”
They are writing specifications:
- inputs and outputs
- preconditions and postconditions
- data models
- failure cases
- performance constraints
- non-functional requirements
We are reinventing documentation, but this time for AI.
The bottleneck never disappeared. It moved upstream. AI doesn’t remove the need for clarity. It demands it.
AI modeled problem-solving by reading all the documents used by successful development projects.
These are text-native, stateful systems. PowerPoint is not.
AI wants specifications. PowerPoint wants summaries.
AI wants boundaries. PowerPoint wants narratives.
AI compounds intelligence. Slides compound meetings.
What Your Execs Will Do Instead
Your executives will ask:
“Can the AI turn this into a better deck?”
PowerPoint is safe and familiar. You will optimize AI for narration, summarization, and post-hoc storytelling. Instead of embedding AI into the operational substrate of the company, they will place it at the reporting layer.
The PowerPoint membrane remains intact. The slide becomes the official truth.
AI is organized around presentation and not performance.
What Winning Teams Will Do
They will not lobotomize AI into PowerPoint. They will not see AI as a translation layer to:
- build ad-hoc dashboards
- map reality to a board level narrative
- summarize complexity
- map execution and complexity into one-page template
AI-Native companies will work where AI was born. They will
- Operate in text-rich environments
- Treat documents as first-class assets
- Let agents participate in workflows
- Minimize slide-based abstraction
- Expose executives directly to operational systems
How Execs Can Set Their Teams Up For Winning with AI
DON’T Ask For an AI Strategy. DO Ask for an Automation Strategy.
Asking for an AI strategy is like asking for an electricity strategy. AI is a utility. Shift the question from the technology to the outcome: what could better automation unlock for your customers and your team?
DON’T Automate Roles. DO Automate Work.
The question shouldn’t be “which jobs disappear?” Accounting wasn’t shrunk by automation. It was leveled up. The headcount stayed; the work changed. Manual data entry turned into real-time intelligence. Ask what tasks are beneath your team’s intelligence, and free them from those first.
DON’T Build AI Islands. DO Use AI to Bridge Silos.
Every company has departments running different systems with different logic. That gap used to require multi-year, multi-million dollar integration projects that rarely delivered. AI can close those gaps faster and cheaper. Aim automation at reducing friction, not at showcases.
DON’T Create AI Tiger Teams. DO Create Space for Exploration.
Tiger teams fix problems. They don’t change culture. If the stakes are too high to experiment, your teams will use AI on the safest, least valuable use cases and call it adoption. Give people room and safety to be beginners again.
DON’T Measure Productivity. DO Measure Value.
Faster-at-the-same-thing is the floor, not the ceiling. Customer expectations have kept pace with every tooling leap engineers have made. The question isn’t can we do this faster? The question is, what can we give our customers now that we couldn’t before?
One group will compound intelligence. The other will compound theater.