A framework for thinking clearly about where AI actually adds value — and where it doesn’t.
Here’s a question worth sitting with: now that AI can write, reason, summarize, and automate, do you still need traditional software at all?
The honest answer is yes — but the more important answer is that the question itself is slightly wrong. The real shift happening right now isn’t software vs. AI. It’s that the interface layer between humans and software is being fundamentally rewritten.
Let me explain what I mean, and why it matters for how you work.
The Old Model vs. The New One
For decades, using software meant learning the software’s logic. You adapted to the machine. You memorized Excel functions, navigated Photoshop’s toolbar, learned keyboard shortcuts. The core skill was: can you operate this tool?
The emerging model is different. You express intent in natural language, and AI figures out which tools to invoke, in what order, to produce the result. The software is still there — often running invisibly in the background — but you no longer need to learn its grammar.
Software isn’t disappearing. Its visibility is decreasing. Like electricity or plumbing, it becomes infrastructure: always present, rarely noticed.
The One Variable That Actually Matters
To understand where AI genuinely helps and where it doesn’t, there’s a single lens that cuts through the noise:
Ambiguity.
Traditional software is a rules executor. It can only handle situations that have been explicitly defined in advance. AI’s core value is handling situations that haven’t been explicitly defined — the fuzzy, the open-ended, the “it depends.”
Ambiguity can appear in three places in any task:
- In the input — you’re not sure exactly what you want, or can’t express it precisely
- In the method — there’s no single correct path from A to B
- In the output — there’s no single right answer
This gives us a simple but powerful framework:
| Ambiguity Level | What It Means | AI’s Role |
|---|---|---|
| Zero ambiguity | Input, method, and output are all defined | Not needed — use software directly |
| Single-point ambiguity | One of the three is unclear | Assists — fills in the gap |
| Multi-point ambiguity | Two or more are unclear | Leads — guides the whole process |
Why This Changes Everything for How You Work
The practical implication is this: AI doesn’t replace categories of software. It replaces the human effort of handling ambiguity within every category.
Take legal work as an example. A law firm’s workflow contains all three levels:
- Billing calculations — zero ambiguity, software handles it
- Filing and compliance management — low ambiguity, some AI assist
- Client intake and case assessment — high ambiguity, AI has real leverage
- Legal research across jurisdictions — high ambiguity, AI genuinely helps
AI doesn’t “take over” legal work. It takes over the ambiguous portions of legal work, in every domain.
The same logic applies to your own work, whatever industry you’re in. The question to ask about any task isn’t “will AI replace this?” It’s: where does the ambiguity live in this task, and what’s it costing me to resolve it manually?
Where AI Is Genuinely Transformative Right Now
The highest-value use cases share a common structure: multi-point ambiguity, repeated frequently, currently resolved by expensive human time.
Learning and comprehension. When you’re entering an unfamiliar domain, you don’t know what you don’t know. You can’t formulate precise questions. There’s no fixed path from “confused” to “competent.” AI excels here not because it has all the answers, but because it can meet you where you are — anchoring new concepts to what you already understand, calibrating depth to your actual goals.
Decision support in complex environments. When the inputs are messy, the variables are many, and experience-based judgment has historically been the only way through — AI can compress the research and synthesis work dramatically, even if the final call still requires human judgment.
Workflow gaps that have never had good tooling. In many industries, the middle layers of work — the coordination, the translation between specialists, the interpretation of ambiguous client requests — have always been handled by experienced humans because software couldn’t touch them. These are exactly the nodes where AI is now creating real leverage.
Where AI Is Overhyped (For Now)
To be fair: AI is often applied in contexts where it adds little value, or where the risks outweigh the benefits.
Zero-ambiguity tasks don’t benefit from AI — they just add latency and a surface area for errors. A script that converts files or calculates payroll doesn’t need a language model in the loop.
High-stakes, low-tolerance domains require caution. The same property that makes AI valuable in ambiguous situations — its ability to generate plausible outputs — is exactly what makes it dangerous when accuracy is non-negotiable. Medical records, legal filings, financial compliance: AI can assist, but every output needs review.
And simple habits die hard. Sometimes the right answer is: use the tool you already know, because the switching cost exceeds the gain.
The Practical Takeaway
Here’s a mental model worth keeping:
Before reaching for AI, ask: where is the ambiguity in this task?
If there’s no ambiguity — if the input is clear, the method is defined, and the output is deterministic — use software. It’s faster, more reliable, and doesn’t hallucinate.
If there’s ambiguity at one or more points — if you’re not sure what you need, or how to get there, or what a good answer even looks like — that’s where AI starts earning its place.
The tools that will matter most over the next few years aren’t the ones that automate everything. They’re the ones that are precisely calibrated to the specific type of ambiguity that costs you the most time and cognitive load in your actual work.
That’s the real question to be asking — not “do I still need software?” but “which of my ambiguities is finally solvable?”
If this framework resonated with you, I’d love to hear how you’re thinking about it in your own workflow. The best product decisions — mine included — come from understanding where the real friction lives.