This week in AI: FedDev's AI money, OpenAI cut-offs and Gmail to-dos

Last week the federal government put another $500 million on the table for businesses that want to use AI, and southern Ontario's share opened for applications the same day. The next day, OpenAI told developers that four of its best-known models will stop working in January. And Gmail started handing paid Workspace users a to-do list it writes from their email.

One of these is money, one is a deadline, and one is a decision you can make before it reaches you.

Ottawa reopened AI adoption funding in southern Ontario

On October 13, the federal government announced an extra $500 million over five years for the Regional Artificial Intelligence Initiative, bringing the program to $700 million in total. It is delivered by the regional development agencies; in southern Ontario, that is FedDev Ontario.

The money goes two ways. One stream helps businesses adopt AI to improve productivity. The other helps Canadian AI companies find customers and scale. The release says the program has funded 218 projects since it began in 2024.

Who it is actually for

The headline number is large. The entry requirements are what decide whether it is for you. For the adoption stream, FedDev's eligibility page says a business must:

• be a for-profit business located and operating in southern Ontario, or a First Nations, Métis or Inuit business;
• be incorporated and registered to do business in Canada or Ontario;
• employ at least 5 full-time equivalent staff and fewer than 500;
• have at least 3 years of operations, shown in financial statements.

The funding is a repayable contribution of $125,000 to $5 million, covering up to 50% of the project. In plain terms, it is a loan for up to half the cost, and the smallest one works out to a project of roughly $250,000. The page gives "acquiring and integrating AI solutions such as predictive analytics, robotics" as examples of what it covers.

So this is not money for a ChatGPT subscription or an AI receptionist. It suits a manufacturer automating quality checks, a distributor rebuilding its forecasting, or a clinic group replacing its scheduling system. A five-person firm with a $20,000 idea should not spend a week on the paperwork.

The two dates that matter

FedDev's program page says applications are open until January 19, 2027, at 4:59 p.m. Eastern. But if your project and its costs start before March 31, 2027, you need to apply by November 24, 2026. That is five weeks away.

What to do

1. Check the four eligibility tests above before anything else. If you miss one, stop here.
2. If you qualify, write one page on the project: what it replaces, what it costs, and how you will measure the result. FedDev will want all three.
3. Decide now whether you are aiming for November 24 or January 19. Repayable funding still has to be repaid, so build the repayment into the plan.

OpenAI is switching off GPT-4o and GPT-4.1 for developers

On October 14, OpenAI posted a deprecation notice: GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini will be removed from its API on January 14, 2027. The API is the connection that software and automations use to talk to OpenAI's models, as opposed to the ChatGPT app you type into. OpenAI recommends moving to a model called gpt-6-luna, and says "fine-tuned versions of these models follow the same shutdown schedule as their underlying base models." A fine-tuned model is one a developer has trained further on your own examples.

MIXED points out that this is 92 days' notice, while OpenAI's own page promises "at least 6 months" for generally available models. It also reports that ChatGPT is not affected, and that fine-tuned models cannot be moved across; they have to be trained again on the new model.

Why it matters to a small business

According to MIXED, OpenAI's own model page still calls GPT-4o "the best model for most tasks", so plenty of small-business tools may be built on it: the email sorter a freelancer set up in 2025, a Zapier or Make step that calls OpenAI directly, the chatbot on your website, a script that writes product descriptions. If any of those names one of the retiring models, it will stop working on January 14.

The upside is cost. MIXED compares OpenAI's list prices in US dollars per million tokens (a token is a chunk of text, roughly part of a word): gpt-6-luna is $0.10 for input and $0.50 for output, against $2.50 and $10 for GPT-4o. For anything you pay for by use, that is a large drop.

The catch is that a new model is not a drop-in replacement.

The catch is that a new model is not a drop-in replacement. OpenAI describes Luna as its "most efficient model for focused, high-volume tasks." The same prompt can come back in a different tone, length or format, and anything downstream that expects a particular format can break quietly. A customer-facing chatbot needs testing on real questions before you switch it, not after.

What to do

1. List every tool or automation that uses OpenAI with your own API key. Your OpenAI account's usage page shows which models you have been calling.
2. Ask whoever built each one: which model does it use, and when will you move it?
3. Test the new model on 20 or so real examples before switching. Leave December for fixes; nobody wants to debug a chatbot over the holidays.

Gmail's AI to-do list reached paid Workspace plans, in the US first

On October 15, Google said Gmail's AI Inbox is rolling out to paid Workspace users. It adds a view that does two things: "suggested to-dos" pulls action items and deadlines out of your email, and "topics to catch up on" summarizes long threads, including decisions and status changes.

It is arriving on the Business Standard and Plus, Enterprise Standard and Plus, and Frontline Plus plans. Domains on Rapid Release began getting it on October 12; Scheduled Release domains start on October 26, and each rollout can take up to 15 days. Google says Starter and Education editions will follow "in the coming months."

Two lines in the post matter more than the rest. It is "currently only available to users in the US with EN-US as their language." And "there is no admin control for this feature."

Why it matters to a small business

For most Canadian businesses, nothing changes this week. But Google tends to widen these rollouts, and when this one arrives, the decision will sit with each person on your team, not with whoever runs your account.

That is worth thinking about before it happens. An AI summary of an inbox can miss the one email that mattered, or turn a customer's "maybe next month" into a confident to-do. And if your staff handle personal information, such as patient details, client files or employee matters, you are still responsible for it under PIPEDA, Canada's federal private-sector privacy law. Google's help page for the feature says it is in beta, works only on the Primary tab, and ignores attachments and encrypted mail. Each user can hide it under Settings, Inbox, by unticking "Show AI Inbox."

What to do

1. Nothing yet if your team is in Canada.
2. If you have staff in the US on a paid plan, tell them it is coming and that a summary is a starting point, not a substitute for reading the thread.
3. Add a line on AI inbox features to your staff AI policy: who may use them, and on which kinds of email.

What we're watching

• How quickly FedDev funds flow. Whether the November 24 date produces approvals before spring, and whether the agency publishes which kinds of adoption projects it backs.
• OpenAI's notice period. Whether OpenAI extends the January date, given the gap between 92 days and its own six-month policy.
• AI Inbox in Canada. When Google extends it beyond US English, and whether it adds an admin setting before it does.

Where to start

1. Check the FedDev eligibility tests against your business today; if you pass, put November 24 in the calendar.
2. Make a list of every tool that calls OpenAI with your key, and ask its builder for a migration date before December.
3. Decide your team's rule for AI email summaries now, while you still have time.

If you want a second pair of eyes on any of these, you can book a short call with Trimotech.