Claude Code and Codex Just Proved AI Has Found Its First Real Business Model — Here's the Service Play That Follows

The Signal

Simon Willison published something important today on HN.

He calculated what his AI usage actually costs — not at the $100/month consumer plan rate he pays, but at API prices that enterprise customers now face.

His numbers, verbatim:

• $1,199.79 for Anthropic Claude Code (30 days at API pricing)
• $980.37 for OpenAI Codex (30 days at API pricing)
• Total: $2,180.16 worth of tokens for $200

That gap is the business opportunity.

What Changed in April 2026

Two things happened in April that closed the door on enterprise discounts:

Anthropic switched their Enterprise plan to $20/seat/month plus API pricing for usage — a change that reportedly occurred in November 2025 and became widely known when existing customers renewed. Reported by The Information on April 14, 2026.

OpenAI made a similar pricing change on April 2, 2026 for Codex, extended to all existing ChatGPT Enterprise plans on April 23, 2026.

On top of that, GPT-5.5 (released April 23rd) is 2x the API price of GPT-5.4. Opus 4.7 (April 16th) is around 1.4x the price of Opus 4.6 when accounting for the new tokenizer.

Companies that signed year-long enterprise deals are now locked into these prices.

Why This Is Product-Market Fit

Willison's argument is sharp: OpenAI had 900 million weekly active users for ChatGPT in February 2026, but only 50 million — 5.6% — were paying consumer subscribers.

Charging $10–$20/month per user is a limited business against $1 trillion in planned infrastructure.

But companies spending $200+/month per user — which is what enterprise API pricing produces — scale the economics dramatically. Coding agents are the first AI use case that burns enough tokens per session to justify this AND delivers measurable value to well-compensated professionals.

OpenAI currently lists 703 open jobs, of which 229 (32.6%) relate to enterprise sales and support. These companies are building enterprise salesforces — because they have a product enterprise customers will pay real money for.

That is product-market fit.

The Service Business This Opens

When enterprise AI costs shift from "low and predictable" to "high and variable," companies need help managing that spend.

Most companies spending $3,000–$20,000/month on LLM APIs have no dedicated ML team, no workflow cost mapping, and no routing logic separating high-stakes tasks from high-volume tasks.

That is your client.

The Audit

Start with a one-time AI spend audit:

1. Map every workflow that calls an LLM — identify the model, prompt pattern, token count, and frequency

2. Classify each workflow: does it require frontier-level reasoning, or can it run on a fast, cheap open-weight model?

3. Project monthly savings from routing changes using real API price deltas

4. Deliver a written architecture recommendation with specific swap paths

Price this at $2,500–$5,000. For a company spending $10,000/month on APIs, saving 40% is a $4,000/month improvement — the audit pays for itself in weeks.

The Retainer

After the audit, the durable revenue is a monthly monitoring retainer:

• Cost tracking with anomaly alerts
• Regression testing when models update — does quality hold on the open-weight paths?
• Quarterly routing review: is the frontier/open split still optimal given new model releases?
• Alert on new pricing changes from major providers — there will be more

Price this at $1,500–$3,000/month. It is an easy yes for any client who just recovered significant monthly spend from the audit.

Who to Target

Three segments:

Fast-moving 2024–2025 AI startups that wired GPT-4 or Claude into everything quickly and never ran cost architecture. Their stack is probably a single provider, a single model tier, and no routing. The audit almost always surfaces savings.

Mid-market product companies that have AI-powered features in production and are seeing API bills grow month-over-month without understanding why.

AI-native agencies running client workflows at scale. They often absorb API margin hits themselves rather than passing costs to clients. The audit reframes this as a profitability problem with a known solution.

The Pitch

Willison's numbers give you a one-line pitch:

"A moderately heavy AI user hits $2,180.16 in API costs for what the consumer plan caps at $200. Enterprise customers now pay API prices. We map where that gap is sitting in your stack and show you exactly how to close it."

That sentence works in cold email. It works in LinkedIn. It works in a 30-second intro at a conference.

The business is real. The timing is now. The evidence is public.