AI selection · a supporting resource

The best AI depends on the job you need done.

Updated October 2026

You do not need to become an AI expert. We start with the job, the information it needs and the people who will use the result.

See how we get started

A practical way to choose

Five questions before choosing a tool.

The job

What job does it need to do?

Name the recurring task, its owner and what a useful result would look like.

The information

What information may it use?

Agree which sources and accounts are allowed, and what must stay private.

The systems

Where does the work happen?

Choose tools that fit the software and processes your team already uses.

The judgment

Where does a person review the work?

Agree approvals, exceptions and what happens when the tool cannot complete the task.

The evidence

What supports the choice—and what is the total cost?

Look beyond a demo or subscription price. Include checking, corrections, retries and running costs.

If you want the detail

How we would make a comparison useful.

The study gives us a research starting point and a proposed testing method. Open the detail that interests you.

How would we compare options?
  • Shortlist: 2–4 suitable options for an agreed workflow; define acceptance before results.
  • Test cases: Include missing information and exceptions. Record model, settings, permitted tools, date and retry limits.
  • Review: Automated checks where possible; two reviewers assess subjective outputs without provider names. Keep disagreements and uncertainty visible.
  • Permission and budget: Agree client-data use, provider access and paid testing separately.
  • Status: Proposed method, not completed tests. The $750 assessment covers one workflow, not unlimited AI-product comparisons.
What can the research tell us?
  • Source: Official provider catalogs, pricing, product pages and terms reviewed October 29, 2026.
  • Useful for: Shortlisting and questions about features, information handling and cost—not proof of business performance.
  • Not measured: No paid model tests, latency trials, client-data experiments, production-success measurements, independent benchmark scores or measured model ranking.
  • Before deciding: Recheck provider documentation and terms; they can change.
What does the total cost include?
  • Running cost: Usage, retries, failures, human checking and corrections, connections and hosting—divided by accepted outcomes.
  • Implementation: Keep one-time costs separate unless an explicit period is used to spread them.
  • Measure the trade-off: A pricier model may or may not reduce review and rework.
  • Calculator: Planning inputs, not predicted success rates.

Source: Mycel’s AI selection guide and supporting study, October 29, 2026. This summary describes the approach; it does not publish the full study or claim independent certification.

Start a conversation

What keeps landing back on your desk?

Bring one recurring task. Let’s talk through a sensible first step.

Book a free 30-minute discovery call
Mycel Group / Meet the founders

Meet Austin & Evan.

At Mycel Group, we bring operating judgment, thoughtful engineering and support for adoption together. We build around the people doing the work.

Austin Barnhart

Founder, Mycel Group

Austin Barnhart

I bring an operator’s perspective to AI. The work comes first; the technology serves it.

Read Austin’s full biography

Operating background

Start with the business.

My background is corporate consulting and operational performance, with a focus on leadership, profitability and the mechanics of how work actually gets done. That experience shapes every question I bring to AI: What is the business trying to improve? Where does the work stall? What would let the team move with more speed and more confidence?

Hands-on with AI

Learn by building.

I’ve spent years working with these tools directly: organizational models, operating dashboards, executive agents and assistants, documentation systems, and the workflows that connect them. That practice is where the judgment comes from. It’s how I know where the tools are reliable, where they need structure, and what it takes to turn a capable model into a dependable part of a business.

How I work

Clarify the business. Then build the workflow.

I work with several leading AI platforms, and the choice follows the work: your purpose, existing systems, budget and information needs.

During assessment, we collect the operating knowledge that informs the proposal. If included in the accepted proposal, we develop and maintain your business canon as part of implementation, with your approval.

From that reference we build task-specific instructions and templates, test them against real examples, and refine together, keeping human judgment at the points where it matters.

Ownership and access are set out in the accepted agreement. Third-party software remains subject to its licenses and ongoing subscriptions. Documentation and training support continued operation; reliability also depends on your systems, access and upkeep. Support and changes follow the agreed scope.

Evan Persson

Technical Co-Founder, Mycel Group

Evan Persson

More than eight years of software engineering and architecture experience, connecting technical decisions to the way a business actually works.

Read Evan’s full biography

I’m Evan Persson, technical co-founder of Mycel Group. I bring more than eight years of software engineering and architecture experience to practical AI implementation, connecting technical decisions to the way a business actually works.

My background spans APIs, data pipelines, secure integrations, and cloud services. I’ve led engineers and owned systems from architecture through deployment and production support. That experience shapes how I approach implementation: reliability matters, ownership needs to be clear, and the work does not end when a system goes live.

At Mycel Group, we bring operating judgment and technical expertise together. We begin with the work a client wants to improve, then establish the information sources, access requirements, connections, and human review points needed to support it. From there, we configure suitable tools, connect them with existing systems, and test against real examples.

As lead developer and software architect for an AI voice-analysis project, I designed a pipeline that converted voice streams into text for language analysis. That work reinforced an important lesson: an AI model is only one part of a useful system. How information moves, how services connect, and how results are checked deserve just as much attention.

I believe clients should understand what we build with them. I document how each setup works, where its limits are, and what the team needs to know to operate it. Accounts, ownership and access are set out in the agreement.

My focus is practical: thoughtful engineering, clear documentation, and AI implementations that fit the people and processes they are meant to serve.

Have a task that keeps coming back?

Book a free 30-minute discovery call