INSIDE TECH QUICK FACTS · JUN 2025 · VOL.02

Are You Struggling to Implement AI?
Using AI and implementing AI are not the same thing.

Most founders use AI tools. Far fewer have embedded AI in their actual product. The gap is not technical — it starts with knowing where to begin.

TL;DR

Using AI in your workflow is not the same as implementing AI in your product. While 78% of businesses use AI in some function, only 34% of software companies have embedded it in their processes. A three-stage framework — Input, Throughput, Output — helps you evaluate every potential AI feature by resource intensity and infrastructure impact. Start with behavioral data. Scale from there.

01

78% of businesses actively use AI in at least one function — but only 34% of software and platform companies have embedded AI in their core business processes. The gap between using and implementing is real, and wide.

02

A non-tech founder built a production customer support agent in under 4 hours using Anthropic's API directly — no Lovable, no Zapier, no developer. The second time would be even faster. Switching costs are dropping fast.

03

"Using AI" ≠ "implementing AI": content tools, social automation, and coding assistants are using AI. Embedding it in your platform's core value delivery — deciding how users interact, how requests are processed, what they experience — is something else entirely.

04

"AI by design" as a decision step: treat AI evaluation as mandatory for every platform improvement — the same way "privacy by design" is built into good product development. Decide first whether AI should be involved, before choosing how.

05

Conversational AI is the most resource-intensive input: every back-and-forth clarification drives up token usage. Starting with behavioral data requires no new user interaction at all — it uses what your platform already knows.

06

Netflix started personalization with purely rule-based behavioral data, only later upgrading to AI-driven personalization. Nothing fancy — just increasingly effective. This is the model: start contained, scale intelligently.

"The critical first step isn't technical. It's the decision to start small, learn fast, and scale intelligently. Implementing AI does not begin with sweeping overhauls — but with targeted, measurable initiatives."

THE AI IMPLEMENTATION STACK — Input · Throughput · Output

For every AI feature you consider, map it across three stages. Your choices compound — a high-personalization input creates high-resource throughput. Start where infrastructure impact is lowest.

I
Input
T
Through­put
O
Output
Industry data — 80%+ of businesses believe AI gives competitive edge; 78% use AI in at least one function
Industry data — only 34% of software and platform companies have embedded AI in their business processes
Anthropic Claude — Sonnet 4, used to build Map Your City customer support agent in under 4 hours
Netflix — behavioral data-first personalization model, later upgraded to AI-driven recommendations
Caroline Vrauwdeunt — "Are You Struggling to Actually Implement AI?", Inside Tech