NetApp
/
What’s in Private AI Certified — Premium
/
When AI Stopped Being an Engineering Project
NetApp
Iterate.ai
NetApp Sellers & Partners
Joint Educational Series
Field Brief · The Adoption Curve

When AI Stopped Being an Engineering Project

Enterprise AI has been through three phases in four years. We are in the third. It is the first one that people who are not engineers are buying.

What you'll learn
  • The three phases enterprise AI has moved through since 2022
  • Why the second phase stalled for reasons that had nothing to do with model quality
  • What changes when the buyer is the person with the problem
The core argument

Every technology that matters starts as something specialists build and ends as something the person with the problem uses directly. AI is partway through that shift. ChatGPT marked the move into phase two. Small models, efficient open-weight releases, and cheaper GPU hardware marked the move into phase three.

November 30, 2022. ChatGPT’s launch moved the market into phase two. Anyone could chat with an AI now, kids included, and professionals brought it to work before policies existed. Running a model behind a firewall was neither cheap nor easy. By 2026 insurers responded: new ISO endorsements let carriers exclude generative-AI claims.
The Three Phases of AI Adoption
2022 and before
Expert Engineer DIY
2023 – 2025
SaaS AI
2026 onward
Complete application platforms
Technology
Bespoke stacks
Assemble from scratch
Custom fine-tuning and hand-rolled retrieval
No standard interfaces
Difficult to maintain
Abstraction layers
APIs to outside platforms
Variable, hard-to-forecast token costs
Data and IP leave your environment
Integration work lands on you
Standardized
Private AI, agents and governance
Set up without a data-science team
Data sovereignty and access control built in
Runs on infrastructure you already own
Buyers
Early adopters
Highly technical teams
Rough edges tolerated. IT largely not involved.
Dev teams
Engineering-led buying
Technical evaluations. Pilots that stay pilots.
Enterprise majority
Business-unit buyers
People with a problem to solve, buying an outcome.
NetApp | Iterate.ai
When AI Stopped Being an Engineering Project · 01 / 02

It has happened many times before

None of this is new. These are three of many. Each time, something only a few could afford became something many could use. Each time, it stopped being a project and became a tool.

Spreadsheets, 1980s

Before them, a finance manager who wanted to model a budget scenario filed a request with IT and waited weeks for a rigid report. Change one assumption, file another request. Spreadsheets put that in the manager’s hands: change a number, see the answer.

Websites, 1990s–2000s

Columbia Business School’s Mattan Griffel traces the cost of starting a company from about $5 million in 2000 — servers, database licences, six-figure salaries — to roughly $500,000 once open source arrived, then $50,000 after cloud computing (Forbes, May 2026). Shopify then took the storefront down to a monthly subscription.

Spaceflight, 2010s–2020s

A kilogram to low Earth orbit cost tens of thousands in the Shuttle era. SpaceX landed its first orbital booster in December 2015, and reuse has since cut that by roughly 90%. Satellite constellations that were financially impossible became ordinary.

What the pattern says about now
  • The capability stops being the hard part. Deciding what to point it at becomes the hard part.
  • The buyer changes. Budget moves from the technical team to the team with the problem.
  • The winners are not the companies with the best specialists. They are the ones whose ordinary staff can use AI.
  • The shift looks obvious afterwards and contested while it is happening.
AIPod Mini is ready to launch chat and agents within 20 minutes

This is what the AIPod Mini is for. NetApp® infrastructure and Iterate’s Generate software arrive pre-integrated, so the work that used to be a quarter of engineering becomes setup. A business team can stand it up and have a working agent the same day — on their own hardware, with their own data.

Ask who has to be in the room

Ask a vendor who has to be in the room. That answer tells you which phase you are buying.

Further reading

On why phase two stalled: The End of “Per Token” Costs and The Rest of the Invoice — what the metered model actually bills you for.

Satya Nadella’s argument that value accrues to whoever owns the learning infrastructure appears in The Reverse Information Paradox, Jul 12, 2026. Iterate.ai has built AI since 2015 and has worked only on private AI since 2023.

About this series

A joint educational series on private AI and the AIPod Mini. NetApp — the governed data-control layer. Iterate.ai — the private intelligence layer.

AIPod Mini
NetApp
Iterate.ai
v1.1 · Aug 11, 2026
NetApp | Iterate.ai
When AI Stopped Being an Engineering Project · 02 / 02