It's Always the Process, Stupid

Unstructured Data and Process Structure

• Debate over the claim that AI is the “first useful tech for unstructured data.”
• Several argue structured vs. unstructured processes long predate AI: checklists, forms, and clear question sets are “structured data,” even without databases.
• Examples: “talk to the vendor” (unstructured) vs. “ask these 10 compliance questions” (structured). Only the latter is reliably automatable.
• Others note many processes cannot practically be fully structured because they:
  • Interface with messy reality or customers.
  • Depend on differently structured systems/teams.
  • Face huge edge-case variability not worth modeling.
• Good design pushes semi-structured “fuzz” to the edges and watches it carefully; AI may make it cheaper to leave more of those edges unstructured.

AI, BPO, and “No Silver Bullet”

• Strong support for the article’s core: automating a bad process just produces bad outcomes faster.
• “There is no AI strategy, only business process optimization” resonates with many, though some argue a good AI strategy becomes BPO.
• Parallel to software: much “tech debt” is really “org debt”; social and technical problems are intertwined. You can’t fix misaligned incentives or hated steps with tooling alone.
• Brooks’ “No Silver Bullet” is cited as still relevant.

Hype, Strategy, and Where AI Actually Helps

• Longstanding pattern: leadership sees new “buzzy-technique” as a cost-cutter, when in fact it needs sustained investment.
• Some say most AI initiatives they see are for customer-facing features and funnels, not internal BPO—often driven by FOMO.
• Others emphasize AI’s real power in handling text and unstructured inputs: routing requests, clarifying ambiguity, replacing low-level playbook work.
• Counterpoint: similar gains might come from simply examining and redesigning the process, with or without AI.

Documentation, Legibility, and Process Design

• Multiple anecdotes where writing down a process exposed that stakeholders disagreed on what was actually happening (e.g., “Step 7” stories).
• Documentation often reveals hidden complexity and becomes a prerequisite for sensible automation (including AI).
• Tension: documenting and “legibilizing” everything can harm culture or flexibility; some explicitly avoid writing things down to dodge being constrained.

People, Process, and Organizational Debt

• Process both protects against lazy/low-effort behavior and risks stifling “rockstars.”
• Suggested compromise: strong default processes for the 80% case plus explicit “escape hatches” and sandboxes for exceptional people/situations.
• Many problems in enterprises are attributed to years of cost-cutting, underinvestment in skilled headcount, and leadership-driven tech debt.

Style, Authorship, and Automation Risks

• Several readers dislike the blog’s “LinkedIn / LLM” tone and suspect AI authorship; the author confirms heavy LLM assistance.
• Some find the HN discussion clearer than the post itself.
• Recurrent theme: AI is best viewed as “automated intelligence” or “accelerated incompetence,” depending on how well the underlying process is designed and governed.