Microsoft backs specialised AI: model choice becomes a work decision

The AI race increasingly includes specialised tools. A useful business choice depends on the task and the cost of reaching an approved result.

By · CEO of Lyntev

Published . Revised

Event date:

Building 92 at Microsoft’s Redmond campus, photographed in May 2016
Building 92 at Microsoft’s Redmond campus, photographed in May 2016. Archive image. · Coolcaesar · CC BY-SA 4.0

Microsoft presented new AI models at Ignite on 17 November. Reuters reported MAI Code 2 for programming and a cybersecurity model intended for a limited group of customers. The performance and price comparisons announced by the company remain supplier claims, rather than findings from an independent test conducted for this report.

Let the task guide the choice

I see a reason here to reconsider how businesses buy AI. The first question need not concern the best-known name. It can concern the work: preparing a reply, classifying documents, changing code or analysing an exception. Each task has different acceptance criteria.

I would write those criteria before comparing products. What makes an answer correct? Which errors are tolerable? Who approves the output? Without those definitions, an impressive demonstration can win the argument without solving the team’s problem.

The cost of an approved result

Usage charges are part of the calculation. In a fictional programming evaluation, I would include review time, necessary corrections and verification of the changed behaviour. A low-priced suggestion requiring several attempts may be less economical than its individual price implies.

I would retain a representative set of tasks chosen before the trial. That allows versions to be compared without selecting only examples where a tool performed well. This is a proposed evaluation method, rather than a test carried out for this article.

More choice needs better criteria

Contractual availability, permissions and data handling also need confirmation. A limited-access announcement does not establish that every customer can start using the product.

My conclusion concerns the purchasing decision: rather than seeking one winner for every job, look for a verifiable match between task, quality and cost. Specialisation becomes an advantage when that match appears in real work.

Sources consulted

  1. Microsoft unveils AI models for coding, cybersecurityReuters, republicação autorizada no Investing.com. Date stated by the source: . Accessed: .
  2. Ignite 2026: Building the frontier ecosystemMicrosoft. Date stated by the source: . Accessed: .

Editorial responsibility

Text and analysis: Wellinton Oliveira, CEO of Lyntev. Sources and the limits of the available information are identified on this page.

Documentary checks of the identified sources, distinguishing announcements, confirmed information and editorial interpretation. No interviews or product tests were carried out for this report.

Wellinton Oliveira

Written by

CEO of Lyntev

Technology with context, original analysis and sources you can consult.