On April 28, 2026, OpenAI and AWS announced that GPT-5.5, Codex, and Managed Agents now run directly inside Amazon Web Services environments. Two days earlier, Microsoft had done the same with Foundry.
That is not a product launch. That is an infrastructure decision.
When the two largest enterprise technology platforms in the world embed AI agents into their core systems in the same week, the question businesses face changes. It stops being "should we explore AI" and becomes "which agents do we deploy first, and who installs them."
The experiment phase is over.
What "Infrastructure" Actually Means for Your Business
Infrastructure is what runs underneath everything else without being noticed. Electricity. Internet. Payroll software. Nobody debates whether to have electricity — they decide who supplies it and what it powers.
AI agents crossed that line in April 2026.
The shift — April 2026
Not because of the OpenAI/AWS deal specifically. That deal is a symptom. The cause is 18 months of model improvement that quietly crossed a reliability threshold. GPT-5.5 and Claude Opus 4.7 (released April 16, with 64.3% accuracy on SWE-Bench Pro, the standard engineering benchmark) can now hold context across complex systems, execute tasks end-to-end, and recover from failure without hand-holding. The outputs are consistent enough to build guarantees around.
That threshold matters more than the press releases.
Before it was crossed, agents were demos. Useful in controlled conditions, fragile in production, expensive to build reliably. We know this because we built them in those conditions and watched them fail in predictable ways. After it was crossed, agents are operational systems. The cost to build drops. The quality floor rises. Guaranteeing outcomes becomes possible in a way it wasn't 12 months ago.
The $29/Month Problem
The same month OpenAI wired into AWS, four new SMB-focused agent platforms launched. Yobi, Pluriza, Agently, Arahi. All targeting the same operational pain points, all priced between $29 and $79 per month.
Self-serve. No-code. One login.
SMB agent platforms — Q2 2026
This is good for the market. It confirms the category. It means the "is AI real?" conversation is over, which is the conversation that was costing everyone the most time.
But here is what those platforms don't have: any understanding of your specific sector. They don't know NDIS pricing band structures, or what triggers a SCHADS breach, or why an after-hours incident at a supported independent living home is categorically different from a support coordination enquiry. They don't know what a coordinator reads before they refer.
Generic agent platforms solve generic problems. A business running on sector-specific compliance requirements needs something built for those requirements. Not adapted from a CRM template.
The $29/month product handles the easy 80%. The remaining 20% is where compliance risk, revenue risk, and relationship risk all live.
Why the Timing Matters More Than the Technology
We spent 13 years inside the NDIS sector before we started building systems for it. We watched providers scale from 3 homes to 30 and absorb six new full-time staff just to handle the coordination overhead that came with growth. We watched referral pipelines stall because a coordinator sent an enquiry on a Friday afternoon and nobody followed up until Tuesday.
The agents we install now would have handled both of those problems. But we couldn't reliably guarantee the outcomes in mid-2024. The models weren't consistent enough. The cost to build reliable pipelines was too high.
That changed in April 2026.
Technology adoption windows
The window between "this technology is reliable enough to guarantee" and "every provider already has it installed" is narrow. It was narrow for practice management software in 2012. Narrow for rostering software in 2018. Narrow now.
The first businesses to move in each of those windows spent the following years competing against operators still doing those jobs manually. That gap compounds.
What Comes Next
The question is not whether AI agents will run inside small businesses. They will. The question is whether the agents running inside your business were installed with an understanding of your specific operation, or bolted on from a subscription platform that has never heard of your sector.
If you want to know which jobs in your operation an agent could take over this quarter, the Operational Diagnostic is a free 30-minute session. We map your current workflows and identify the two or three jobs leaking the most time and revenue. No pitch. No proposal unless the numbers make sense.