QuarterlyNovember 1, 20268 min read

HeitechSoft Quarterly: Fall 2026

August to October 2026

In 30 seconds

This quarter, we launched PaiSLEY, the AI hiring tool we first built for our own hiring.
Our lead story explains how it reads each résumé against the job while a person makes every decision, and how it sorted more than 300 applications for one developer co-op role before anyone opened the dashboard.
Our lab notes cover giving AI assistants only the data each person may see, planning for the day an AI provider goes down, and checking every AI-generated figure against its source.
Our wins include a sales system for a language-training company that is now in daily use, and our support for AI builders at the Guelph Engineering Competition.
Five industry developments follow, including new federal AI adoption funding and AI drafting invoices in QuickBooks Online.

Introduction

This quarter we launched PaiSLEY, the AI hiring tool we first built for our own hiring, and we now run our own hiring through it. The surprise came from using it. With more than 300 applications for a single role, the résumé often turned out to be a weak guide to what a candidate can actually do, and a candidate's final score could land a long way from the first match against the résumé, in either direction. I no longer rule anyone out on the résumé alone, and I'd suggest you don't either.

— Tom Heiber, CEO

PaiSLEY is live, and we run our own hiring through it

On September 23 we launched PaiSLEY at paisleyhr.ai. It reads each résumé against the job and explains how the candidate matches, with strengths and concerns. While a candidate is still applying, it asks follow-up questions to uncover experience the résumé leaves out. The AI recommends and a person decides: every shortlist or rejection asks for confirmation, and every status change is recorded in an audit trail.

An HR, payroll and staffing firm asked us to close the gaps against the applicant-tracking system it uses, and on October 1 we deployed its own PaiSLEY instance. Its team told us "we've been looking for years, and those functionality things are not out there."

We use it ourselves. After we sponsored the Guelph Engineering Competition, our posting for an AI-first developer co-op role drew more than 300 applications in about a week, and PaiSLEY had sorted them all before anyone opened the dashboard, so our review started with the top ten.

What it means for you: a firm of 20 to 200 people without an HR department can have every application read and ranked the day it arrives, while the person who knows the role makes every decision.

Lab notes

Letting AI assistants into our business data, but only as far as each person may see

With AI coding agents we built our own operations platform in about two months, covering prospects, clients, projects, time, invoices and profitability. In October we connected it to Claude and ChatGPT so our team can ask questions about the business in plain language. An automated check proves that each of the assistants' 49 read tools returns exactly what the person's own screen would show, and that no money figure reaches a role without financial access. Before you connect an AI assistant to your own data, test it the same way.

The day our main AI provider went down

Nearly all of our code is written by AI coding agents, working overnight from written specs and tickets. Our build rules say an agent's own report that its work is done is not evidence, so a separate verifier agent checks the work before it is merged. On September 3 Claude was down, and every agent on our builds stopped. We moved the work to OpenAI's Codex, which held up well. If your business comes to depend on one AI tool, keep a second one tested and your instructions written down.

A founder found that AI had invented their research figures

An industrial hardware start-up we support found that figures AI tools had given it two years earlier, and that it had relied on in its research, were made up. That is the flaw we warn about most: these tools can state invented facts with complete confidence. Our advice was to use the research modes every major AI provider now offers, which link claims to sources, and to open those sources. Any AI figure in your plan, funding application or investor deck needs a source you have checked yourself.

Wins

A language-training company now runs its sales on the AI-assisted system we built

The AI-assisted sales system we built for a language-training company went into daily use: since September 14 its sales lead has run all sales work through it. Whenever a note, email or meeting is logged, the AI suggests next steps and tags the contact, applying tags it is confident about and leaving the rest for a person to confirm. A second phase, now in the client's testing, sorts incoming emails, while a person always makes schedule changes. The client plans to switch over fully by the end of November.

Backing the next AI builders in Waterloo Region and Guelph

We sponsored the University of Guelph's Guelph Engineering Competition on October 3 and 4, where 120 student competitors took part, and our team attended the closing ceremony. In September we sponsored "AI Agents for Everyone", a Waterloo Tech Week event in Kitchener that drew about 70 to 80 people, where our team introduced HeitechSoft. On September 3 Tom was interviewed as an AI expert on the Dr Linda Show, a live stream for leaders of small and medium-sized businesses.

By the numbers

95%
of the code changes to our own software this quarter (1,396 of 1,467) were made with an AI coding agent.
We recommend AI coding tools from daily use: almost all of our own software changes now come from them.
424
pull requests on two client builds this quarter, from our AI coding-agent pipeline.
Building in small, separate changes means each one can be reviewed and tested before it reaches the client's system.
300+
applications for one AI-first developer co-op posting in about a week, all sorted by PaiSLEY before we opened the dashboard.
A small team can start its review with the top of the list rather than the full pile.
33
releases of PaiSLEY in three weeks of October.
Frequent releases show how quickly the product is changing as we use it on our own hiring.

AI for mid-market

StatCan: one in three firms with 20 to 99 staff now plans to use AI, double last year

Statistics Canada reported that planned AI use more than doubled among businesses with 20 to 99 employees, to 32.8% from 15.0%, and that the most common planned uses were data analytics, large language models and chatbots. For a business your size, roughly one in three of your peers is now planning an AI project, up from about one in seven a year earlier. Plans vary widely by sector, so compare yourself with firms in your own industry before you set a budget.

Source: Statistics Canada

Ottawa adds $500 million to regional AI adoption funding; southern Ontario applications open to January 19

FedDev Ontario announced an extra $500 million over five years for the Regional Artificial Intelligence Initiative, bringing federal funding to $700 million to help businesses adopt AI. In southern Ontario, applications are accepted until January 19, 2027, and projects with eligible costs before March 31, 2027 must be submitted by November 24, 2026. For a business your size, this is the main federal funding route for an AI project here, but call FedDev Ontario before you budget, because no cost-share or maximum amounts are given.

Source: FedDev Ontario (Government of Canada)

Ottawa funds 10,000 co-funded placements to help businesses adopt AI

The federal government announced up to $162 million over five years for Mitacs to fund 10,000 co-funded work placements, in which students, recent graduates and postdoctoral fellows work with Canadian businesses on AI adoption. For a firm without in-house AI skills, a placement is a practical way to get a readiness assessment, a roadmap or a first implementation done. Mitacs lists a $5,000 business contribution toward a $10,000 placement, so budget some of your own supervisor's time too, because the intern will not know your processes.

Source: Innovation, Science and Economic Development Canada

Microsoft rebuilds Copilot around chat, delegated tasks, a no-code app builder and an always-on agent

Microsoft announced a new Copilot in three parts: Home for chat and delegated tasks, Code for building small apps, trackers and dashboards from plain-language descriptions, and Autopilot, an agent that works toward a goal it is given. Everyday Copilot use in Word, Excel, Outlook and Teams stays on a fixed per-user licence, while the more autonomous parts run on usage-based billing with admin controls over spending. For a business your size, a tracker that someone builds by hand each week is a sensible first trial, but set a spending limit first and remember that most of this is early access or preview as of November 1.

Source: Official Microsoft Blog

QuickBooks Online in Canada adds AI that drafts invoices and runs routine tasks, in beta

Intuit's October update for Canadian QuickBooks Online adds AI that drafts invoices from saved documents such as statements of work, sets up overdue reminders and forecasts cash flow, though the AI is in beta with monthly limits. If you already run QuickBooks Online, these tools arrive in software your bookkeeper uses daily, so a trial costs staff time rather than a new contract. Start with overdue-invoice reminders or the cash flow forecast, and have your bookkeeper check every result, since Intuit warns the AI can make mistakes and additional fees may apply.

Source: Intuit QuickBooks Canada

Written by

Tom Heiber, Founder & CEO, HeitechSoft

Tom founded HeitechSoft in 2004. Since 2022 he has run it as a working “AI company experiment”: applying AI across the business and sharing what actually holds up in practice.

LinkedIn About HeitechSoft Talk to us

If you want to see what this could mean for 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.