Flat Numbers in the Wider Economy #

Tech giants show strong numbers from selling hardware and models. The rest of the economy looks completely different.

Torsten Slok published an analysis on Apollo about this gap. For the S&P 493, revenue per worker is flat. Profit margins are flat. Broad productivity growth shows no movement.

Venture firms spend heavily on tokens. Standard enterprise companies pull back. Buying tokens does not lift the bottom line. It just adds an expensive cost center.

Calculation Is Not Judgment #

Large language models are good calculators. They scan text and generate code quickly.

Business delivery runs on judgment, not pure calculation. Judgment depends on context. It requires understanding culture, trade-offs, and unwritten client expectations.

Models do not understand trust or accountability. They do not carry reputational risk when a production system breaks. When errors reach customers, the company pays the bill.

Automation Still Needs Human Validation #

We see this directly in enterprise software engineering. Automated AI agents can scan pull requests, check style rules, and summarize changes. That part is fine.

Merging stays human. A real person must verify the user interface. A real person must run manual accessibility checks with screen readers. AI output without direct human verification drops quality fast.

Relying on raw model output creates false confidence. Productivity comes from stable pipelines, clear component interfaces, and human verification where it makes sense.

Grounded Delivery Beats Model Hype #

Buying the newest model tier does not solve architecture problems. It does not replace talking directly with people.

Use automation for repetitive pipeline steps. Keep your delivery pipelines clean with automated linting and sanity tests. Leave architectural direction, scope defense, and final system reviews to experienced engineers.

The financial payoff will stay missing until organizations focus on pragmatic delivery instead of model hype.

My Conclusion: Use AI as a Tool, Not a Holy Grail #

We should focus on using AI as a tool, just as we do with every other technology, instead of trying to reinvent the wheel again and again.

Product companies should stay focused on building reliable products. Consultants should stick to what they do well: understanding client context, architecture, and delivery. Creative agencies are still needed too, because real brand character and creative direction do not come out of a prompt. If everyone tries to rebuild everything from scratch around AI, costs quickly get out of reach.

Take proven open-source tools, reuse them, and adjust them where needed. Stop treating every wrapper or custom setup as the holy grail that can be sold at the highest price.