Actually Useful #27
AI tools that actually work at work | Saturday, May 30, 2026 | 4 min read
The decisions being made this week about who funds AI research and where it gets built will matter more to your team’s daily tools than any product launch — and most teams aren’t paying attention to them yet.
Tool of the Issue
Ava 2.0 — an AI sales rep that claims to run outbound on its own
Best for: Revenue ops and sales teams evaluating whether AI can actually replace BDR headcount
Ava 2.0 bills itself as an AI Business Development Rep that runs outbound sales autonomously. That’s a specific, testable claim — and it’s one that sales ops teams have been waiting years for someone to actually deliver on.
Here’s what’s genuinely interesting about it: “autonomous outbound” is either one of the most useful things AI can do for a revenue team, or it’s a well-dressed email sequencer with a chatbot bolted on. The difference matters enormously. Real autonomy means it handles prospecting, personalization, follow-up, and objection handling without a human in the loop. Most tools that use this language are doing the first two and calling it done.
The honest limitation here is that Ava 2.0 doesn’t make it easy to verify the specifics. What does “autonomously” actually cover? Does it send emails without human review? Does it adapt based on replies? What happens when a prospect says something unusual? These aren’t nitpicky questions — they’re the difference between a tool that saves your team 10 hours a week and one that sends awkward emails to your best prospects.
The other thing worth checking before you sign up: compliance. Autonomous outbound touches CAN-SPAM, GDPR, and increasingly aggressive spam filters. If the tool doesn’t have clear answers on how it handles opt-outs and sending limits, that’s a gap worth probing before you hand it your contact list.
Bottom line: Worth a demo if you’re in sales ops and have a clear benchmark in mind. Go in with specific questions about what “autonomous” includes, and ask for case studies with measurable results. Don’t let the demo be a slide deck.
Openstatus MCP Health Checker — testing AI agent connections before they break in production
Best for: Technical ops teams and developers deploying AI agents inside business workflows
This one is narrower than most tools we cover, but if your team is building or deploying AI agents, it solves a real problem that’s easy to overlook until something breaks.
MCP — Model Context Protocol — is the emerging standard that lets AI agents connect to external tools and data sources. Think of it as the plumbing that lets an AI assistant actually do things: pull from your CRM, update a spreadsheet, check a calendar. As more teams move from “AI that answers questions” to “AI that takes actions,” MCP connections become critical infrastructure.
The Openstatus MCP Health Checker tests those connections the way a real AI client would, not just with a basic ping. That distinction matters. A ping tells you the server is on. It doesn’t tell you whether the AI agent can actually authenticate, retrieve data, or handle edge cases correctly. Integration bugs tend to hide in exactly the gap between “server is responding” and “agent is working.”
The audience for this is genuinely narrow — you need to be building or managing MCP servers for it to be relevant. But if that’s your team, this is the kind of testing tool that catches problems before users do, which is always worth having.
Bottom line: Not for everyone, but if your team is deploying AI agents that connect to external systems, add this to your pre-launch checklist. It’s the difference between testing that something is alive and testing that it actually works.
Workflow Win
The regulation gap: why the rules being written today may not apply to the tools you’ll use tomorrow
Something worth paying attention to is happening simultaneously in several different places, and the pieces only make sense when you look at them together.
Colorado’s Governor Jared Polis signed a law this week limiting how chatbots can interact with children — a reasonable, well-intentioned safety measure. At roughly the same time, Maharashtra’s Chief Minister Devendra Fadnavis announced that the state will provide innovators access to 2,000 GPUs with government backing. China is explicitly accelerating its embodied AI development with what it’s describing as full-chain manufacturing support. And in the US, proposed new federal funding rules would give the government the ability to cancel research grants at any time, effectively making peer-reviewed AI safety research contingent on political approval.
These aren’t unrelated stories. They’re the same story.
The core problem is this: Western governments are writing rules for AI while simultaneously defunding the research that would make those rules enforceable and the domestic industry that would have to follow them.
Colorado’s chatbot law only applies to US-based companies operating in Colorado. It doesn’t apply to models trained and deployed from regions where 2,000-GPU government programs are actively recruiting innovators. The companies that will build the next generation of AI agents — the ones your team will likely be using in two or three years — are increasingly being built in places where those guardrails don’t exist, partly because the investment and infrastructure are moving there.
This isn’t an argument against child safety laws. Restricting AI access to minors is a legitimate public health measure and worth doing regardless of what’s happening elsewhere. But regulation without investment is a partial answer. If the US is making research funding politically contingent while eliminating peer review, and other regions are writing checks with fewer strings attached, the practical result is that the safety research gets done where the money is — and the safety standards get written where the products are built.
Here’s what this means if you’re a team lead or ops manager making AI tool decisions right now: the tools you’ll be evaluating in 2027 and 2028 will increasingly come from vendors operating under different regulatory frameworks than the ones your legal team is currently tracking. That’s not a reason to panic, but it is a reason to ask harder questions about where your AI vendors are based, where their models are trained, and what compliance standards actually apply to them — not just which ones they claim to follow.
Skip This One
Perplexity as your “trustworthy” research tool — not yet
The citation-first AI search engine is facing a lawsuit from CNN over alleged scraping without permission
Perplexity built its reputation on doing citations right — showing sources, attributing claims, being the more transparent alternative to AI tools that just generate answers with no paper trail. For research-heavy teams, that was a genuinely compelling reason to choose it over other options.
The problem is that CNN has filed suit alleging that Perplexity scraped content without permission. If the core value proposition is “we cite our sources,” and the legal challenge is “you used those sources without authorization,” that’s not a peripheral issue — it’s a direct hit to the thing that made the product worth choosing in the first place.
Citation and permission are different things. Showing where content came from doesn’t mean you had the right to use it. For teams that chose Perplexity specifically because it felt like the more ethical option, this is worth watching before you deepen your reliance on it.
When to revisit: Once there’s legal clarity on the CNN case and a clear statement from Perplexity on how they’re handling publisher agreements going forward.
The most useful question you can ask about any AI tool right now isn’t “what can it do?” — it’s “where was it built, and what rules does it actually have to follow?”
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.