AI implementation, for automation and agents.
Upskailed’s AI implementation service builds AI agents, automated workflows, and internal AI tools for the tasks your team repeats, inside your own tools and accounts.
Jonathon Bloomfield, a PECB Certified ISO/IEC 42001 Lead Implementer, leads every implementation project.
Building an AI product? See the AI stack review, which examines the models, prompts, settings, and software your product uses.
What Upskailed builds,
after checking whether a tool you pay for can do the task.
Automated workflows for tasks your team repeats
Internal AI tools
The AI stack review for startups,
a second opinion on how your product uses AI.
A written review of how your product uses AI, with a ranked list of changes, about two weeks after you provide access. A call to discuss the findings follows. Making the changes is optional and scoped separately.
Which model the product uses for each task, with open-weight and smaller model families considered.
What the product sends to a model on each request, and what it could cache.
Tests that protect quality when models change.
What your logs retain, and how much the product depends on one vendor.
Where model spending goes, and which changes would lower it.
Read-only access to the code that calls the models, the prompts and settings, a usage or billing export, and a de-identified log sample.
A security audit or penetration test, legal or privacy advice, investor due diligence, or a savings guarantee.
Start with a short request,
and plan the rest together.
A short request form
What you need, who it’s for, and when.
A free introductory call
A conversation about your situation, and whether working together makes sense.
A written scope and proposal
The work, the deliverables, and the price, in writing. Nothing starts until it’s approved.
Development in stages, then handover
You review each stage before the next, and a small group of staff tests the system before everyone uses it. It’s then documented and handed over, with 30 days of support.

Work with Jonathon Bloomfield
Founder of Upskailed, ISO/IEC 42001 Lead Implementer and Lead Auditor
A former high school math teacher, Jonathon has spent thousands of hours working with AI, and teaches what has worked for him to professionals, founders, and leadership teams.
Ontario Certified Teacher · Claude Certified Architect, Foundations and Professional (Anthropic)
About JonathonQuestions before your AI implementation project
Sometimes, yes. Every engagement starts with a review of whether a tool your organization currently pays for can do the job. If it can, Jonathon sets it up, and anything custom covers only what the tool can’t do.
It depends on the scope: the systems, data sources, and review points involved. You get a written scope with options after a free call, and each option shows what it would cost to run.
Most first implementation projects take about six to ten weeks from the written scope: one to two weeks to agree the task and gather the information it needs, one to three to set it up, and a pilot with a small group of staff before everyone uses it. From you, it needs one person responsible for the project, a few hours a week from them during setup and the pilot, and access to the accounts involved.
You do. Everything is built in your own tools and accounts, then documented and handed over so you or your staff can continue running it.
Reducing dependence on one vendor is part of every scope. You own the system and it runs in your accounts, and the operating guide names each model and what to do if it changes. Where a system does depend on one vendor, the written scope names it, so you can decide whether that’s acceptable before anything starts.
Models from Anthropic, OpenAI and Google, plus open-weight and smaller model families, through direct APIs or cloud hosts, or AI gateways. For open-weight models, the scope names each model’s licence terms and where data is processed.
30 days of support are included. After that, optional ongoing support covers monitoring, fixes and small changes, month to month with 30 days’ notice. New features require separate agreements.
The operating guide says what to do when a model changes, and ongoing support includes checks when one is retired or repriced. Vendor and model prices change often, so they’re dated whenever they’re quoted.
Anything that makes clinical, legal, financial, or eligibility decisions about people, crisis services, or tools aimed at children and youth without added safeguards. For tasks that support these decisions, the system can prepare information for the person who makes the decision.
Everything runs in your own accounts and is documented in an operating guide, so your staff or another developer can maintain it. You own the system, so nothing has to be handed back.
At review points agreed in the scope, before anything is sent or changed. An agent does the steps you’d otherwise click through, then waits for a person to approve. Assistants answer from your own sources.
The system runs in your own accounts, and the operating guide states what each tool logs and where it stores data, so you can decide what to keep.
Start with the task you repeat most,
in the tools you use now.
Share the task, the tools involved, and who approves, and Jonathon replies within three business days to set up a free call.