AI Adoption Rates Starting to Flatten Out

Interpreting the “flattening” claim

• Many commenters argue the headline is overstated or wrong given the charts: small-firm adoption is clearly rising; large-firm adoption shows recent declines but not a clear long‑term trend.
• Others counter that several consecutive months of decline, especially in larger firms, does start to look like a real trend, absent evidence of a transient shock.
• Some suggest a more accurate framing would be “stagnation” or “plateau,” especially for big companies.

Definitions and data quality

• Strong confusion and disagreement over “adoption” vs “adoption rate”:
  – Some read “rate” as a time derivative; others as “percentage of firms using AI.”
  – Title vs axes vs text are seen as internally inconsistent.
• Criticisms of the charts: no y‑axis label, misleading use of “rate,” 3‑month moving average hidden in a footnote, and missing grid lines.
• The two datasets (Census vs Ramp) differ by ~3x, raising questions about representativeness and methodology.
• Census question wording (“use AI in producing goods or services”) may exclude a lot of incidental or exploratory use, making absolute levels hard to interpret.

Enterprise vs small business dynamics

• Small firms (1–4 employees) show the cleanest, steadily rising line; some see this as the leading indicator, with larger firms to follow later.
• Others note that large-firm adoption appears to have fallen, which is concerning given AI company valuations and capex plans.
• Speculation that grassroots use (developers, tiny businesses) grows while mid‑/large‑company managers slow or resist adoption if they feel threatened.

Bubble, valuations, and macro outlook

• Several commenters see classic bubble dynamics: huge capex and valuations assuming exponential revenue growth, but only modest or stalling adoption.
• Expectation from some that most AI startups will be wiped out or acquired cheaply; long‑term value may still be huge but concentrated in a few players after a shakeout.
• Others say flattening would be a normal point on a technology S‑curve, not necessarily an “AI winter.”

Use cases, productivity, and developer experience

• Mixed experiences on coding:
  – Some see LLMs as strong force multipliers for boilerplate, glue code, and minor tasks.
  – Others find quality unreliable, edge cases missed, and codebases turned into “slop” that’s hard to maintain.
• Several developers report skill atrophy, loss of deep understanding, and reduced joy in programming when relying heavily on AI; a few intentionally cycle “no‑AI weeks” to keep skills sharp or have quit AI tools entirely.
• Debate over whether non‑adopters will be unemployable in ~5 years vs. whether deep engineering skill (independent of AI) will remain the scarcest and most valuable asset.

Agents, UX, and mainstream adoption

• Strong skepticism that autonomous “agents” are actually in real, unsupervised production use; most usage is interactive co‑pilot style.
• Many note the average person doesn’t know what to do with LLMs; value will rise as AI is embedded into familiar applications and workflows, hiding complexity.
• Growing frustration with hallucinations and inconsistent answers (e.g., between different models) erodes trust and may dampen further direct-chat adoption.