What Can You Do With Your QuickBooks When You Can Just Talk To It?
Published: May 22, 2026
The Practical AI Show, hosted live every Friday at 11am CT by Olga Pechnenko and Chris Pearson, helps business operators cut through the AI hype cycle. We track what's actually shipping, who's getting funded, which predictions are landing, and how non-developers can use AI in real businesses right now. Subscribe at @TheRealPearsonified.
On May 18, 2026, a jury took under two hours to unanimously dismiss every one of Elon Musk's claims against OpenAI. The dismissal was on statute-of-limitations grounds — Musk filed too late. OpenAI, Sam Altman, Greg Brockman, and Microsoft were all cleared. No damages awarded. No forced nonprofit reversal. Musk called it a "calendar technicality" and is appealing. OpenAI filed for a $1 trillion IPO four days later.
MCP (Model Context Protocol) is the protocol that lets AI agents reach into real systems and execute actions on your behalf. Olga's plain-English explanation: "MCP is like an API protocol that literally gives AI models hands. Your model is a brain just sitting there. MCP is the hands that come into your system and execute things on your behalf." On May 20, 2026, the NSA's AI Security Center published a Cybersecurity Information Sheet flagging four systemic risks: agents blindly trusting tools, tools firing erroneously, unchecked tasks passing between agents, and vulnerabilities in the data packaging layer. Independent benchmarking (MCPTox) found attack success rates above 60% across 45 live MCP servers.
Three things in one week: Uber's CTO revealed they burned through their entire 2026 AI budget by April. Microsoft cancelled its own internal Claude Code licenses because the cost couldn't be justified — developers are being moved to GitHub Copilot CLI by June 30. GitHub Copilot itself is dropping flat-rate plans entirely on June 1, moving to usage-based, token-metered billing. Combined with Anthropic's June 15 split (chat stays in subscription, agentic work moves to API), the pattern is clear: outcome-based pricing is the destination.
Skip the QuickBooks developer API — it requires a 45-minute privacy assessment built for software companies, not individual operators. Use the Claude Cowork QuickBooks connector instead. Open Cowork, create a folder for your project, create a new project pointed at that folder, go to Customize → Connectors, connect QuickBooks (read-only by default, write requires approval). Then write a prompt explaining your business and asking Claude to build a weekly dashboard. Olga's exact prompt is in the show.
The framework is identical for any software where you piece reports together manually every week. Sales reports, lead pipelines, project status, ad performance — the workflow is the same. Connect the source, write a prompt explaining your business, ask Claude to recommend what to measure, then build the dashboard. As Olga said on the show: "What is the one thing that you don't like to do? How can you make this an agent for the week?"
An architecture pivot from chatbots to agents at the operating-system level. Gemini Spark is a 24/7 cloud agent that works while your phone is physically locked. Gemini Intelligence provides screen-aware automation across apps. Gemini Omni Flash offers free conversational video editing on YouTube Shorts. Gmail Live lets you query your inbox by voice. Google's AI Mode crossed one billion monthly active users. The smartphone is becoming a "digital chief of staff" that executes objectives without you ever clicking around an app.
A plain-text map of your business that AI agents can read — like robots.txt for the AI era. Google added an Agentic Browsing check to Chrome Lighthouse that audits domains for an llms.txt file. The same week, Google's Search team published guidance saying the file is optional. Mixed messaging, but the direction is unmistakable: make your website machine-readable.
An X user posted a Claude Monet painting (Water Lilies, 1915) claiming it was AI-generated and asked critics to describe what made it "inferior to a real Monet." 6.7 million views later, people had picked apart the brushstrokes, called it "soulless garbage," and trashed a literal masterpiece — purely because they thought a machine made it. Then the mic drop: it was the real Monet. The lesson for business owners: anti-AI bias is reflexive and irrational. Don't talk about using AI — just build the thing.
"Daddy's home. He wants his credit card back."
— Olga, framing the end of the free AI era
"This is as close to a fairy tale as I've ever heard. I guess that's what happens when there's almost nine hundred billion dollars in the IPO on the table."
— Olga, on the Musk verdict
"MCP is like an API protocol that literally gives AI models hands."
— Olga, explaining what MCP actually does
"It's like giving AI your corporate credit card and master key to your server room."
— Chris, on MCP without guardrails
"Going to QuickBooks and running any kind of report is so dinosaur-like."
— Olga, on the pain that drove the demo
"It's literally better than QuickBooks."
— Chris, reacting to the dashboard Claude built
"One tiny little fleet startup can now take on somebody like Microsoft in the next two years. It's real."
— Chris, on the generational opportunity for builders without technical debt
"What is the one thing that you don't like to do? How can you make this an agent for the week?"
— Olga, the universal call to action
Olga opened the show with the line that traveled across X and LinkedIn all week: "Daddy's home. He wants his credit card back." The crazy unlimited AI subscription era is going away, especially in the corporate world. Anthropic and OpenAI are both preparing for IPOs, which means Wall Street accountability — quarterly earnings, real profit margins, pricing discipline pushed down to every user.
This show traces three signals that all landed in the same seven days: Uber burning its 2026 AI budget by April, Microsoft cancelling its own Claude Code licenses, and GitHub Copilot ending flat-rate plans. The pattern is unmistakable. Compute is not free. The buffet is closed. The meter is running.
On Monday, May 18, the jury in Musk v. OpenAI took under two hours to unanimously dismiss every one of Musk's claims. The reasoning: statute of limitations. Musk waited too long to file. The judge agreed. OpenAI, Sam Altman, Greg Brockman, and Microsoft all walked away clean.
Olga's prediction from Episode 41 hit cleanly: "I don't think Elon Musk is going to win. There's too much money at stake." But the sub-prediction — that OpenAI would pay some fines into a nonprofit arm — missed. OpenAI paid nothing. Walked away with a clean slate.
And not by coincidence: OpenAI filed for a $1 trillion IPO four days later. Targets a $60 billion raise and a September 2026 public debut. Goldman Sachs and Morgan Stanley are leading. The Musk verdict directly cleared the structural ambiguity that was making institutional investors nervous. "This is as close to a fairy tale as I've ever heard," Olga said. "I guess that's what happens when there's almost $900 billion in the IPO on the table."
Sources: NPR · NBC News · CNBC on the IPO filing
Google's I/O 2026 was not a model release. It was an architecture pivot toward agents as the new operating system.
Chris on what this means: "AI is going to be the thing that's primarily using software in the future. It's not going to be humans clicking around on interfaces. It communicates directly with the data layer of the software and it just gets things done. It goes beyond the UI." The implication for businesses: your beautifully designed app interface stops mattering when the OS-level agent never opens your app. Your UI is becoming an API endpoint for an AI agent to hit.
Source: Google Blog · Gemini Intelligence · Engadget I/O roundup
Google added an Agentic Browsing audit category to Chrome Lighthouse. One check: does your site have an llms.txt file. That's a plain-text summary at your domain root that AI tools can read to understand your business. Robots.txt, but for ChatGPT, Perplexity, and Gemini.
But Google can't agree with itself. The Lighthouse team added the check. The same week, Google's Search team published a mythbusting guide saying the file is completely optional. "Google's llms.txt guidance depends on which product you ask," Search Engine Journal wrote.
Mixed messaging aside, the direction is unmistakable. Your website is now being evaluated on whether machines can understand it. Not just humans. Not just search crawlers. Agents.
Sources: Search Engine Land · Search Engine Journal
On May 20, the NSA's AI Security Center published a Cybersecurity Information Sheet on the Model Context Protocol (MCP). Government-grade scrutiny for a protocol that was a developer buzzword a year ago.
Olga's plain-English explanation: "MCP is like an API protocol that literally gives AI models hands. Your model is a brain just sitting there. MCP is the hands that come into your system and execute things on your behalf."
Chris's framing on the risk: "It's like giving AI your corporate credit card and master key to your server room. Convenient for automation. But without hard-coded access controls, you've automated a massive security breach."
The NSA flagged four systemic risks:
Independent research backs the concern: the MCPTox benchmark tested 45 live MCP servers and found attack success rates above 60%. A bad actor doesn't need to hack a firewall. Just a prompt injection hidden in an innocent customer query.
Source: NSA AI Security Center · MCP Security Design Considerations
Three corporate moves landed in one week. All pointing the same direction.
Uber burned through its entire 2026 AI budget by the end of April. A full year of funding gone in four months. The CTO revealed the number internally. Compute costs ran far past the spreadsheet's predictions.
Microsoft cancelled its own internal Claude Code licenses. The company that owns a chunk of OpenAI couldn't justify paying Anthropic for a competing coding tool. Developers are being moved to GitHub Copilot CLI by June 30.
GitHub Copilot is dropping flat-rate plans entirely on June 1. Strictly usage-based, token-metered billing. The industry-standard flat-rate AI subscription is ending.
The mechanism behind it: agentic token burn. A human types a prompt and burns maybe 500 tokens — a single linear transaction. An autonomous agent operates in a recursive loop: reads a codebase, writes a function, runs a test, hits an error, rewrites, tests again. A single request can trigger 50 automated calls. Bad code (legacy systems, technical debt) makes the loops longer. Costs spiral.
Olga: "Compute is not free. It's our human thing that we got used to with SaaS over time. Nothing is free."
Sources: Windows Central on Microsoft · Startup Fortune on Uber · GitHub Blog on Copilot billing
A viral X post landed mid-week. An artist posted a Claude Monet painting (Water Lilies, 1915) claiming it was AI-generated, asking critics to describe what made it "inferior to a real Monet."
6.7 million views. Comments piled on. Brushstrokes critiqued. Called "soulless garbage." A literal masterpiece by one of history's greatest painters trashed — solely because people assumed a machine made it.
Then the mic drop. It was the real Monet. Every critic who'd torn it apart was reacting to "AI made this," not to the painting.
Olga's takeaway for business owners: "AI bias is becoming huge. So when you're building a business, just solve the problems and let it kind of work for itself. Don't talk about using AI. Just build the thing." Chris added: "Art is supposed to be a direct expression of human ingenuity. So if AI is doing it, it's like an affront to humanity. People get overly emotional about this. But this same attitude bleeds over into other areas."
The two top AI executives at WordPress both exited in the same week. James LePage, the original founder of WPAI (acquired by Automattic in December 2024 and made WordPress's Chief AI Officer) left at the end of his 18-month retention term. The second-in-command exited at his 12-month minimum the day after.
Chris's read: "AI's never been hotter. The directions things are going are becoming more clear. The opportunities are unfolding. Now is the time to strike if you are in AI anything. You know why you leave a venture right now? Because it's going nowhere. AI just killed WordPress. If you've got systems that rely on this, you've got a death clock. You've got to move your systems to something that is fleet and agile and ready for the future."
Anthropic, meanwhile, is sucking up talent like a magnet. Andrej Karpathy joined this week. The pattern: top AI talent leaves stalled ventures and consolidates at the companies building what's next.
This week's deep dive: a non-developer (Olga) connecting her real-world recruiting firm's QuickBooks to Claude — without building a developer app, without writing any code.
The first attempt failed. The QuickBooks developer API required a 45-minute privacy assessment built for app companies serving hundreds of clients. Not a path for a single business owner. Olga walked away after an hour.
The pivot: the Claude Cowork QuickBooks connector. Open Cowork. Create a folder for the project. Create a new project pointed at that folder. Go to Customize → Connectors. Connect QuickBooks (read-only by default, write requires explicit approval). Connect Gmail. Connect Calendar.
The honest caveat: Cowork's email connector can only create drafts — it can't send. The workaround Olga used: schedule the report to refresh weekly and have Cowork drop a calendar event reminding her to check the dashboard at 8am Monday.
Olga's prompt (worth stealing):
"I run a small business and I want a simple weekly cash dashboard built from my QuickBooks data. I'm not technical. I think in outcomes, not spreadsheets. First, before building anything, based on my business, tell me the five most important cash numbers a small business owner my size should look at every week. Keep recommendations short and plain language. Then connect to my QuickBooks account and build the dashboard as a single self-contained HTML page I can view right here."
Claude came back with five numbers: cash on hand, cash in this week, cash out this week, net this week, money owed to me (overdue). Then built the dashboard. One delightful HTML page replaced five separate QuickBooks reports.
Chris's reaction: "That something that in QuickBooks only lives on separate reports that you have to look at separately and kind of put the data together in your head — you can have that presented in a single report that's more delightful, more visually exciting, easier to read, more shareable. It's literally better than QuickBooks."
The unlock isn't just data display. It's conversation. "Is 8 weeks of cushion good or bad for a business like mine?" Claude answers based on your actual business context. "Draft a chase email for my most overdue invoice." Done. The static report becomes a thinking partner.
Chris's strongest prediction of the episode:
"The operating assumption for the last three decades has been that you don't want to compete in tech against the big boys. If you're going to compete against Microsoft, you're stupid. That has been true. Telling you right now, that's all changing. One tiny little fleet startup can now take on somebody like Microsoft in the next two years. It's real. Somebody starting from the ground up without all this systemic waste can move slowly now to build the right system, and move so fast later that even Microsoft couldn't catch you with all the money in the world."
The thesis: legacy companies have codebases too messy to refactor at AI scale. The cost of cleaning up their technical debt with AI is higher than rebuilding from scratch. Fleet operators without that debt are positioned to outrun them. Microsoft's cancellation of Claude Code is the symptom. The diagnosis is structural.
The universalizer. Olga's takeaway for the half of the audience that doesn't use QuickBooks:
"If you're not using QuickBooks, forget QuickBooks. You just saw how to make an agent without being a developer who understands API. If you want to generate 50 fresh leads every Monday — go talk to Cowork, figure out the prompt, figure out what the report should look like, set up a scheduled task that gets delivered to you at 8am Monday. Whatever in the business is kind of a pain in the ass for you to do — start thinking how you can use this workflow to build an agent. What is the one thing that you don't like to do? How can you make this an agent for the week?"
The framework is identical for any software where you piece reports together manually. Sales. Marketing performance. Project status. Inventory. Ad spend. Customer support. Pick the pain. Connect the source. Write a prompt. Build the dashboard you wish existed.
Week 25 of the Practical AI funding tracker. The headline: AI funding hit $2.3 billion across 64 companies — down 76% from $9.45 billion the week before. AI share dropped from 60% to 28% of all venture funding.
Top 5 AI rounds this week:
The trend: AI capital migrated from the model layer into verticals. Retail. Healthcare. Energy. Legal. Workforce. Security. "The biggest week for AI headlines was the smallest week for AI funding."
Full Week 25 report: practical-ai-funding-may-21.pages.dev. 25-week tracker: practical-ai-funding.pages.dev.
Chris's close: "These companies have messes to clean up that are so large that I say they cannot be cleaned up. You have to hit the reset button. That's not what these companies do. That is a real existential crisis for them. What that really means is fleet operators have an opportunity — a generational opportunity. It's the time of the little guy. It's the time of the person who's agile, the company that is agile, and really focused on being aligned with the future. It is worth it if you're this type of builder to go all in for alignment with the new realities right now."
Olga's close: "My realization this week is that we follow the 30,000 feet of AI. But the gap between people who are cloud-code experts and pros, and the rest who are still trying to figure out how to even start — the gap is getting so big. How do we keep building technologically advanced, but also make sure we don't leave our customers behind?"
That's the challenge for every business operating in the new AI age. The tools are here. The protocols are paving. The pricing is correcting. The opportunity is generational. The hard part is closing the gap — between what's possible and what your customer actually understands.
Subscribe to the Practical AI Show on YouTube. New episodes every Friday at 11am CT. Hosted by Olga Pechnenko and Chris Pearson.