Actually Useful #21
AI tools that actually work at work | Saturday, May 16, 2026 | 4 min read
Your cloud bill is going up, your bank account might soon talk to ChatGPT, and the electricity powering all of it is getting expensive fast — here’s what’s actually useful to know this week.
Tool of the Issue
ChatGPT wants to help you manage your money — and it can now connect to your bank account
Best for: Finance-conscious ops managers and individual contributors tired of spreadsheet hell
OpenAI has launched a personal finance feature inside ChatGPT that lets you connect your bank accounts directly. The idea is straightforward: instead of exporting CSVs, building pivot tables, or paying for a separate budgeting app, you ask ChatGPT questions about your spending and it answers with your actual data.
This is genuinely useful territory. Connecting financial accounts to an AI assistant isn’t new — tools like Copilot for Finance have done this in the enterprise space — but OpenAI bringing it to a consumer-facing product means a much larger group of people can try it without a procurement process or IT approval.
What works well here is the use case clarity. Personal finance is one of those areas where people know they should be paying more attention but the friction of doing so is high. If ChatGPT can lower that friction, that’s real value.
The honest trade-offs: there are no published details yet on the security and compliance framework for handling sensitive financial data. That’s not a small thing. Regulatory requirements around financial data handling are real, and until OpenAI publishes specifics, enterprise teams should treat this as a personal tool rather than a team or company one. There’s also no independent testing of how accurate the analysis actually is — early access claims and real-world performance don’t always match.
Bottom line: Worth trying for personal use if you’ve been curious. Hold off on connecting business accounts until the compliance picture is clearer.
YouTube now scans for AI deepfakes of your face — and it’s available to all adults
Best for: HR and communications teams worried about synthetic media impersonation
YouTube has expanded its AI deepfake detection tool to all users 18 and older. The feature is designed to detect when someone’s likeness has been used in synthetic media without their consent — essentially, if someone makes a deepfake video using your face and uploads it to YouTube, the system is supposed to catch it.
For HR and communications teams, this matters more than it might seem. Executive impersonation via synthetic video is a documented and growing problem. Fraudulent videos using a company leader’s likeness to spread misinformation, solicit payments, or damage reputation are no longer hypothetical. Having a platform-level detection layer is a meaningful defensive step.
The limitations are worth knowing. YouTube hasn’t disclosed detection accuracy rates, so there’s no public data on false positives or false negatives. It’s also not fully clear whether the tool catches audio deepfakes or is primarily video-focused — “likeness detection” is the language used, which leaves that question open. And there are reasonable questions about what scanning for your face at scale means for user privacy that haven’t been fully addressed.
Bottom line: This is a useful protective layer that costs you nothing to have active. It won’t catch everything, but it’s better than no detection at all. Communications teams should know it exists.
Workflow Win
Your AI roadmap has a power problem — and it’s not the one you’re thinking about
Most conversations about AI costs focus on software: subscription tiers, API pricing, seat licenses. But there’s a physical constraint building underneath all of it that’s starting to show up in real numbers.
Power prices on America’s largest electricity grid are up 76%. Lake Tahoe — not exactly a known tech hub — is facing energy price spikes tied directly to AI data center demand in the region. NextEra Energy, one of the largest U.S. utilities, is reportedly in talks to acquire Dominion in a deal explicitly connected to growing power demand from data centers. These aren’t separate stories. They’re the same story.
The core problem is a mismatch: AI companies are building bigger systems faster than the physical grid can support them, and the cost of that gap is starting to flow downstream to everyone who uses cloud infrastructure.
Here’s how this connects to your team’s actual work. When data center operators pay more for electricity, cloud providers absorb some of that cost and pass the rest along. It’s not always visible as a line item — it shows up as price adjustments, tier restructuring, or reduced discounts at renewal. If your team has been seeing cloud costs creep up without a clear explanation, energy is part of the answer.
There’s a counter-argument worth taking seriously: utilities are actively responding. Acquisitions like the reported NextEra-Dominion deal are specifically about building capacity to meet this demand. Renewable energy and battery storage are scaling faster than most infrastructure timelines historically have. This may be a painful transition period rather than a permanent structural crisis.
But “transition period” can mean two to five years. For teams building AI roadmaps right now, that’s relevant. If your plan assumes that compute costs stay flat or decrease, you’re working with an assumption that deserves a second look. Energy availability and electricity pricing are now legitimate inputs to technology strategy — not just concerns for the facilities team.
Here’s what this means if you’re an ops or finance lead: ask your cloud vendors how they’re thinking about energy costs in their pricing models over the next 18 months. It’s a reasonable question, and the answer will tell you something about how seriously they’ve thought it through.
Skip This One
VinFast’s AI-era expansion playbook — not a model worth following
A cautionary tale about building infrastructure before the demand is there to fill it
VinFast raised significant capital, built manufacturing capacity, and moved fast on an ambitious timeline. The theory was that scale would create a competitive advantage. The reality: $6.9 billion in debt obligations, forced divestiture of core manufacturing assets, and a public pivot from a growth story to a “path toward profitability” narrative — which usually means years of losses ahead.
This isn’t a story about EVs specifically. It’s a story about what happens when a company builds for a market size that doesn’t yet exist, funded by capital that assumed the demand would arrive on schedule.
The lesson for non-technical teams evaluating vendors: a large funding round is not the same as a sustainable business. If a vendor’s pitch is heavy on raised capital and ambitious capacity, and light on unit economics and retention data, that’s worth probing. VinFast thought it could out-manufacture Tesla by building factories first. The market didn’t cooperate.
When to revisit: if VinFast demonstrates two or three consecutive quarters of profitable operations at meaningful volume, the story changes. Until then, the restructuring is still playing out.
If you’re connecting a bank account to any AI tool this week — ChatGPT or otherwise — what’s the one financial question you’d actually want it to answer first?
Tested by AI. Approved by people who just want to get work done.
Know someone who said “we should be using AI more” in a meeting last week? Forward this.
On Substack? A quick Recommend helps more teams find us — takes 2 seconds.
Want the skeptic’s take on what’s real vs. hype?
Overfitted cuts through AI noise with data and zero patience for buzzwords. Mon/Wed/Fri.
Build your own?
The Autonomous AI Newsletter Pipeline Guide: GumRoad or Dodo
How we source, classify, synthesize, and publish at scale.
From the toolbox: Noise Cancelling Headphones (budget)