QuarterlyFebruary 1, 20268 min read

HeitechSoft Quarterly: Winter 2026

November 2025 to January 2026

This quarter, AI became how we build, and owners came to us for the strategy. Over the holidays we built our own AI hiring tool with an AI coding agent and used it to screen the real applicants for our Sales Manager search, with a person making every decision. In the lab, an AI coding agent had a shipping prototype running 15 minutes after its specification, though the client’s payback sums said no; a manufacturer’s CEO assumed our AI-made video used real people; and an AI coding agent built a community group’s door-prize app. Our wins: an aerospace engineering firm asked for a monthly AI session through 2026, a Kitchener accounting firm went from a two-hour AI session to a phased plan, and a language-training company began an Efficiency Edge Discovery. We also cover six AI developments, from Ontario’s new AI-disclosure rule for job postings to Microsoft’s lower Copilot price.

Introduction

This quarter AI stopped being only something we advise on and became how we build. The moment that surprised me most came on January 6: a hiring tool I had built over the holidays with an AI coding agent was screening real applicants for our own Sales Manager role. In the same weeks, owners asked us where to start, and what they wanted was a strategy, a policy and a plan they can follow one step at a time. One quick prototype also reminded us that the payback sums still come first.

— Tom Heiber, CEO

We built our own AI hiring tool over the holidays, then used it on our own hiring

Over the holiday break we built our own applicant tracking system with an AI coding agent, working from a written specification. For each applicant it reads the résumé against the job, asks three follow-up questions about gaps or unclear points, transcribes an optional short video introduction, and gives a match score with its reasoning for a person to review.

Safeguards are built in. The questions never touch protected grounds such as age, gender, religion, health or family status, the assessments use gender-neutral language, applicants never see their score, and a simple check with no AI screens out uploads that aren't résumés.

From January 6 it screened the real applicants for our own Sales Manager search, 145 applications across seven postings by January 23. An outside HR consultant compared her own assessment with its scores and found herself "pretty aligned with the system", while noting it still needs human oversight. A person made every decision.

For your business, a screener shaped to your own hiring process no longer needs a software budget. Since January 1, Ontario employers with 25 or more staff must say in the job posting when AI screens applicants, and our consultant added that line to our postings before screening began.

Lab notes

A working prototype 15 minutes after the spec, and a client who did the maths

A Cambridge machinery builder asked where AI could help, and we started with its spare-parts shipping. We watched its shipping coordinator re-enter the same details four to five times across separate systems. The same day we wrote a specification, and about 15 minutes after it was saved an AI coding agent had a working prototype running, with orders, carrier choice and packing lists. The client liked the dashboard but declined: at its shipping volume, the payback wasn't there. A prototype is cheap now, so do the payback sums on your own volumes first.

An AI-made video the client's CEO assumed had real people in it

An Ontario composites manufacturer preparing to launch a new product came to us in November. Before our first meeting, an AI built a fuller picture of the business from its documents. In December we delivered a short explainer video on recycling, made almost entirely with AI, and its CEO said he had assumed we hired real people. We then built its new website with 45 AI-generated images, converted to a format that cut their size by 91.8%. If your best material sits in PDFs, as this company's did, Google and ChatGPT can't find it.

An AI coding agent built a community group's door-prize app

As an in-kind sponsor, we built a door-prize app for Locally Connected Community Hub, a local business networking group, with an AI coding agent. Attendees check in with a QR code, prizes are matched to their interests, and volunteers verify each winner's claim code. The agent handled the design as well as the code, and in a live test we fixed bugs by handing it the error messages. A small app for one event, the kind no vendor would build for you, now takes days of direction rather than a budget.

Wins

An aerospace engineering firm asked us for a monthly AI session through 2026

An aerospace engineering services firm had people already using ChatGPT, Claude, Gemini and coding assistants with no company policy, while handling confidential client data. We recommended an AI strategy first, with a policy that follows from it as a living document, plus hands-on training for its engineers. In January we ran a working session on AI-written software tests, plain-English database questions and Microsoft Copilot. Afterwards the firm asked us for a 90-minute AI session every month for the rest of 2026. If your staff already use public AI tools, start with the policy.

From a two-hour AI session to a phased plan in three weeks

The owner of a Kitchener accounting firm came to our two-hour AI Explained session in early January. When we visited two weeks later, his team was already using AI deep research on new clients, and we found the same data typed into several systems. In late January we sent a discovery summary and a phased proposal: one view of each client across the practice software, with AI preparing client research and meeting briefs. Each phase pays off on its own, so the firm commits to one step at a time.

Mapping a language-training company before adding any AI

A Toronto language-training company signed an Efficiency Edge Discovery with us in December. Its onboarding ran across several separate tools, and much of the know-how sat with a few people. Rather than starting with a tool, we interviewed each role and gathered its process documents into an AI notebook. For sales we showed an AI-assisted "ideal day" that opens each morning with a prioritised list of follow-ups, and the sales lead said it left her feeling inspired. The discovery continues, with a roadmap and pilot options to follow.

Teaching AI strategy one step at a time, and mentoring graduate students

On January 22 Tom spoke at a local business networking lunch on a step-by-step strategy for bringing AI into a business, using our hiring tool as the worked example. Tom is also a mentor with Humber Polytechnic's International Graduate School, which thanked its mentors during Mentoring Month in January. Owners want a plan they can follow one step at a time, not a list of tools.

By the numbers

15
minutes from a saved written specification to a working prototype built by an AI coding agent. A clear written specification is what turns an idea into a working prototype quickly, which makes testing an idea cheap.
223 of 487
code changes in our software projects this quarter made by AI coding agents, counting only those the tools signed. Much of our own software work now runs through AI coding agents, so we test them on real projects before we recommend them to you.
8
new software projects started this quarter, from our AI hiring tool to a community group's door-prize app. Each new project is a small tool that a business like yours might never have had built under a traditional budget.
90+
client and team meetings this quarter recorded and summarised by AI note-takers. AI note-takers leave a written record of client conversations that everyone can check, without anyone typing it up afterwards.
6
AI and custom-software proposals in three months. Owners are asking for a plan and a clear payback before they commit to a build, and our proposals start there.

AI for mid-market

Ontario job postings must now disclose AI screening

Since January 1, Ontario employers with 25 or more employees must state in public job postings whether AI is used to screen, assess or select applicants, along with new rules on pay ranges and experience requirements. For a 20 to 200 person business, any applicant tracking system, job board or recruiter that ranks or filters résumés with AI means your postings need the disclosure, so ask vendors exactly what their tools do.

Source: Littler

Microsoft cuts Copilot to US$21 a month for firms with 300 or fewer staff

Microsoft 365 Copilot Business now costs US$21 per user per month, down from $30, for organisations with 300 or fewer employees. For a 20 to 200 person firm, a paid pilot for a handful of staff is easier to justify, but confirm the Canadian price with your Microsoft partner, and if you already pay the old price per seat, contact Microsoft to switch.

Source: Computerworld

QuickBooks AI agents arrive for Canadian customers

Intuit has made a team of AI agents available to QuickBooks customers in Canada, including an Accounting Agent for categorizing transactions and a Finance Agent for profit and loss and cash flow summaries. For a 20 to 200 person business, check which agents your plan includes, start with transaction categorization, and have your bookkeeper review the work before month-end, since output is English-only at launch.

Source: Intuit QuickBooks Canada

Budget 2025 backs sovereign AI computing and tracks how firms use AI

Budget 2025 proposes $925.6 million over five years for sovereign public AI infrastructure and a Statistics Canada program to measure how organisations use AI. For a 20 to 200 person business the practical items are in the tax rules: ask your accountant how immediate expensing applies to the computers and network gear you plan to buy, and whether the higher SR&ED limit covers in-house software or process development.

Source: Government of Canada, Budget 2025

Microsoft 365 Business Basic and Standard prices rise July 1, 2026

On July 1, 2026, Microsoft raises list prices for Business Basic and Business Standard, citing AI features such as Copilot Chat in Word, Excel and Outlook. Customers renewing before then can keep current pricing on a new multi-year term, so check your renewal date with your partner, and test the free Copilot Chat before buying paid Copilot seats.

Source: Office Watch

KPMG: 93% of larger Canadian firms use AI, but only 2% report a return

KPMG Canada's survey of 753 leaders at companies with $50 million or more in revenue found 93% use AI in some form, but only 2% see a return on their generative AI investments. If payback is proving hard at your size, you are in good company: pick one or two tasks, time them today, and measure again after 90 days.

Source: KPMG Canada

If one of these stories sounds like your business, book a 20-minute conversation at https://heitechsoft.com/contact.

Get the next issue by email

Each quarter we share what we built, what we learned and the AI news that matters to mid-sized businesses.