Why AI Won't Speed Up Your Processes: The Real Bottleneck

A recent Hacker News post by Frederick Van Brabant titled "I don't think AI will make your processes go faster" has sparked a heated debate. With 232 points and 167 comments, the discussion reveals deep unease about how AI is changing—or not changing—software development. The core argument: AI might speed up coding, but the real bottleneck is everything else. This matters for anyone building software.

What the HN Post Argues

Van Brabant's post contends that AI, particularly large language models, primarily accelerates the development phase—writing code, generating documentation, and so on. But the overall process from idea to deployment involves many steps: requirements gathering, design reviews, legal sign-offs, integration testing, and deployment. These steps often depend on human judgment, communication, and organizational coordination. As he puts it, "AI won't make your Gantt chart shrink" because the longest poles in the tent are not the coding tasks.

The article draws a parallel to the early days of high-level programming languages: compilers made coding faster, but didn't eliminate the need for specification and testing. Similarly, AI might compress the execution part, but the planning and validation overhead remain. Teams should be skeptical of claims that AI will drastically accelerate delivery.

Why the Discussion Exploded

The post struck a nerve because it challenges the prevailing hype that AI is about to revolutionize developer productivity. Comments quickly pushed back, but also agreed in unexpected ways.

One commenter argued the premise is too narrow: "This article assumes that AI only has an impact on the development phase which is certainly not true. It can speed up every part of the step. Including ideation, legal, documentation, development, and deployment." They pointed to AI for ideation, document generation, deployment manifests, and cloud knowledge.

Another comment highlighted the perennial bottleneck: "I think when LLMs first came out people thought they could just say something like, 'Make a Facebook clone'. But now we're realizing we need to be more exact with our requirements and define things better. That has always been the bottle neck in software." This resonates with anyone who has interpreted vague specs like "Get data and give it to the user."

A more cynical take: "It is true that AI doesn't generate the correct programs every time but sadly it has become a common practice to involve AI in every aspect of software engineering, and it is true that it made software engineers become product managers and their work has become to debug and test the entire codebase which adds more frustration."

"The developers being able to drain a one year long work queue in an afternoon is meaningless if the rest of the business cannot absorb the effects of that work in the same timeframe."

This last point—from comment 4—captures the systemic issue: acceleration in one part just shifts the bottleneck elsewhere.

My Perspective

Van Brabant is right that AI won't magically shrink the entire process, but I think the conversation often misses a deeper truth: AI forces us to confront the fuzziness of software requirements. The real promise of AI is not just faster coding, but better specification. The same LLMs that generate code can also help clarify vague ideas, generate edge cases, and produce test plans.

However, the current tooling is still primitive. We ask "write a function that does X" but rarely leverage AI to first help us define X precisely. The bottleneck is not just process—it's the quality of the input. If we treat AI as a code generator without improving how we articulate needs, we'll just generate bugs faster.

AI can also impact non-development phases. Imagine a product manager using an LLM to generate a detailed requirements document from a one-paragraph brief, then a developer refining it with technical constraints. The same model could produce test cases, deployment scripts, and monitoring alerts. The potential is there, but it requires rethinking the entire workflow, not just plugging AI into one step.

What This Means for Builders

If you're adopting AI tools, consider these concrete changes:

1. Start with requirements, not code. Before asking AI to write a line, engineer the prompt to include context, constraints, and desired outcomes. This is the new "design document."

2. Invest in prompt templates for each phase. For example, have a template for generating user stories, one for architecture decisions, and one for deployment plans. This standardizes output and reduces ambiguity.

3. Use AI to automate validation. Instead of manually checking requirements against code, use AI to generate tests that verify the behavior. This closes the loop faster.

Here's a simple example of a prompt that clarifies requirements before code:

## Goal: Generate a function to calculate shipping cost

**Context:** E-commerce site, US only, orders >$50 get free shipping, otherwise $5.99 flat rate. Weight-based surcharge: if weight > 5 lbs, add $10.

**Slippery details:**
- What if order total = $50 exactly? (Shipping free)
- What if weight is unknown? (Assume baseline 2 lbs)
- Should surcharge apply before or after free shipping? (After)

**Output:** First, list any further questions. Then provide the function in Python.

This approach makes the human think about edge cases, and the AI can generate the code with those assumptions explicit.

4. Measure lead time, not coding time. Track the time from idea to deployment. If AI speeds up coding but requirements gathering still takes weeks, the overall improvement is marginal. Focus on the longest waits.

5. Restructure the team. Some organizations have separate product managers and developers. AI can blur these roles—engineers may need to own more of the specification process. That's both a challenge and an opportunity.

Should You Care?

If you're a solo developer or a small team shipping prototypes, AI can still massively accelerate your output because you own the entire process. But in larger organizations with silos and approval cycles, the bottleneck is organizational, not technical. Before betting on AI to shrink your project timeline, map the entire value stream. If the slowest steps involve human coordination, AI won't help much—yet. The real leverage comes from redesigning those steps with AI in mind.

---

Originally published on Hacker News. Read the full article by Frederick Van Brabant: I don't think AI will make your processes go faster.

For more on prompt engineering, see OpenAI's Prompt Engineering Guide. For process mapping, check out The Lean Startup.