Model Vendors Just Built Implementation Arms

The hard part of AI adoption has moved from "which model do we choose" to "how do we get it into the work". The companies selling the models have just shown us that shift.

Part 1: Read It, Then Consider What It Means for Your Company

The companies that supply AI models are launching separate businesses to support implementation. Both OpenAI and Anthropic have started operations that place AI engineers inside customers' workplaces. For Anthropic, plans were reported in July 2026 for an enterprise AI services company backed by investors including Blackstone.

Enterprise demand sits behind this. OpenAI states that as of 2026 its enterprise revenue exceeds 40 percent of the total, and is on track to match consumer revenue by year end. If supplying the model were enough, there would be no need to build an implementation business at this scale.

The meaning of the move is plain. The side selling the models has concluded that models alone do not reach customer results.

For Japanese companies in the Philippines, the implication is that the stage of agonising over "which model" is ending. Performance gaps between models have narrowed and prices have fallen. What creates a difference is how you divide your own work into stages and decide which stages to hand over.

And that design cannot be done by someone who arrives from outside and reads a document. They have to be in the workplace and see the actual steps. That is why the vendors are sending engineers.

There is one more thing to read here. The fact that this kind of business is viable means many companies are stuck in adoption. Contracts signed and models unused is a widespread condition.

Related: Contracting for AI While Prices Keep Falling: Designing for the Switch explains this in detail.

Part 2: Key Terms Explained (for Executives)

Implementation support is support that does not end at delivering a tool. It runs from deciding what the tool is for through to the point where the work actually turns in the field.

Use case refers to a unit of work. "Use AI in accounting" is too large a unit. Narrow it to "match invoice contents against purchase orders" and you can judge whether it can be handed over.

Embedding matters especially at a local subsidiary. With a few Japanese expatriates and a majority of local staff, the people actually running the system every day are the local staff. Making it operable in English, and making sure it does not stop when an expatriate is reassigned, are precisely the conditions for it to stick.

Related: OpenAI Moves Into Hands-On Deployment: An AI Adoption Strategy for Japanese Companies in the Philippines explains this in detail.

Part 3: Applying This to Your Own Company

The vendors starting implementation businesses does not mean a local subsidiary can hire them. It becomes a matter of finding a partner that fits your scale.

So there are things to settle in-house first.

First, the line between what you hand over and what you keep.

I once outsourced a system implementation entirely, left the necessary conditions vague, and ended up with something that ran but could not be used. When it went well, I decided the initial design and the criteria for judgement myself, and left the implementation and the details of daily operation to the contractor.

This line does not change whoever the partner is. If you hand the design and the judgement criteria outside, your company retains no axis on which to evaluate what comes back.

Second, an inventory of the work.

Before receiving support, write out your current steps. Write out which stage has whom doing what, and where a person moves data by hand. Without this list, a consultation produces only the other side's general advice.

Third, deciding what to measure, in advance.

The effect of adoption cannot be measured if you try to measure it afterwards. Decide what you will count first. Total hours saved tends to become an overstatement, so choose something countable and hard to argue with, such as "days from data to first draft".

I was once commissioned by a woman in her 30s running an online English school to build a large system in Next.js. Email sending and receiving, schedule management, customer management, management of multiple instructors, class schedules, evaluation management for both students and instructors, creation and management of teaching materials, and semi-automatic generation of blog articles — everything the operation needed, built as one whole. The budget was over 10 million yen and development took roughly three to five months.

Building at that scale in one pass was possible because we decided at the outset what connected to what. Had we not first organised the structure where bookings, instructors, materials and evaluations interlock, we would have been rebuilding partway through.

Related: What a Major IT Services Firm's AI Partnership Reveals: The Three Things That Close the Gap Between a Capable Model and an Actual Result | Case Study for Japanese Companies in the Philippines explains this in detail.

Part 4: Common Failure Patterns (What Not to Do)

Wrong 1: Starting from a comparison of models

Starting by researching which model is superior only consumes time. Performance gaps have narrowed and prices are moving. Switching has also become easier than it used to be.

Better: We divided the work into stages first, chose one stage to hand over, and then selected a model suited to that stage.

Wrong 2: Rolling out company-wide at once

Contracting several tools simultaneously and starting in every department leads to exhaustion before anyone becomes proficient, and things revert.

Better: We ran first-line enquiry handling only for one month, and expanded to the next stage after it was turning.

Wrong 3: Designing it to be run by expatriates alone

A system operated only in Japanese stops when an expatriate is reassigned. A form that local staff cannot take over has not been embedded.

Better: We produced procedures that can be operated in English, in a form local staff can amend themselves.

Wrong 4: Leaving measurement until later

You end up unable to answer when asked how much effect the adoption produced, because there is no starting point for comparison.

Better: Before adoption, we counted and recorded the time required for the target work and the number of reworks.

Three Tips for Getting Value

Tip 1: Have the design done by someone who has seen the workplace

The vendors are placing engineers on site because design cannot be done from documents alone. When you ask for support, make it a condition that whoever designs has seen your workplace.

Tip 2: Put the judgement criteria in writing and keep them in-house

Write down what is handed over and what a person judges. Without this document, a change of staff means starting the conversation from the beginning.

Tip 3: Start with a unit you can walk back from

For the first one, choose a stage where stopping would not halt the business. Start with core operations and you cannot reverse if it does not work.

Bonus: How to Make Use of PH AI Works

PH AI Works supports Japanese companies considering entry into the Philippines, and Japanese business professionals who already have a local base, in putting AI and technology to work. On this topic of implementation support, we can help in particular with the following.

• Taking inventory of your current work steps and organising, together, the line between stages to hand over and stages a person judges
• Producing procedures that can be operated in English, and designing a handover that does not stop when an expatriate is reassigned
• Deciding what to measure before adoption, so the effect can be explained afterwards

We offer a 30-minute free consultation for Japanese companies in the Philippines. It is fine to consult us before you have decided to adopt anything.

Sources

• OpenAI, "The next phase of enterprise AI" (OpenAI official; announcement on the next stage of its enterprise business)
• Anthropic Newsroom (Anthropic's official news listing; announcements on enterprise business and partnerships)
• Anthropic (developer of Claude; product and enterprise service information)

Note that the substance and timing of these companies' businesses will shift. This article reflects what has been announced and reported as of August 2026. Check each company's official announcements before making contractual decisions.