I Gave Claude One API Key. My CRM Built Its Own Dashboard. | Practical AI Ep 39

Practical AI: Episode 39

I Thought APIs Were For Developers. I Was Wrong. Watch What I Built.

Watch on YouTube

Published: May 1, 2026

TL;DR

  • The Musk-OpenAI trial opened Monday. Sam, Brockman, and Nadella are all on the witness list. The headline $134 billion is not a payday for Musk—he is asking for it to be redirected to OpenAI’s charitable arm. Reuters coverage.
  • OpenAI rewrote three of the biggest deals in AI in seven days. Microsoft’s IP license is no longer exclusive. Google put $10 billion into Anthropic with $30 billion more tied to milestones. Three weeks earlier OpenAI quietly bought TBPN.
  • OpenAI shipped GPT-5.5 under the codename Spud on April 23. It leads or ties Claude Opus 4.7 and Gemini on agentic benchmarks. Pro tier API pricing doubled. Launch announcement.
  • One evening with Claude and a Loxo API key replaced 13 years of manual reporting. A live, color-coded dashboard the team can compete on. No code. Red to green in three days.
  • Funding spiked to $24.18 billion across 86 companies. 82.2% of every venture dollar went to AI—the highest share ever recorded on this show. 7 of the top 16 rounds went to robots, drones, and defense.

Table of Contents

About This Show

Practical AI is a weekly live show (Fridays 11am CT) hosted by Olga Pechnenko and Chris Pearson that cuts through AI hype to deliver news, trends, and hands-on tips for builders and founders. Unlike technical AI podcasts, Practical AI focuses on business applications and ROI—what actually works, what’s hype, and what you can implement Monday morning.

What You’ll Gain

  • Understand what the Musk-OpenAI trial is actually about. Most newsletters reported “Musk wants $134 billion.” That is not the story.
  • Learn what changed in OpenAI’s Microsoft deal. Azure exclusivity is over. The IP license is non-exclusive through 2032 and OpenAI now has multi-cloud access.
  • Discover what an API actually is and why it matters now. APIs have existed since 1993. Until AI showed up, they were a developer-only privilege. AI is the translator that finally opens them up.
  • See a real before-and-after of one operator’s data. A 13-year-old recruiting firm replaced its manual sheet with a live dashboard in one evening. We show the screenshots and the prompts.
  • Gain a practical playbook for what to build first with AI plus API. A daily scoreboard for the metrics that matter, plus drill-down reports on the activities you want to investigate.

Biggest Takeaway to Implement: Pick one tool you already pay for. Ask if it has an API. Paste the credentials into Claude Code and ask it to read the docs and tell you what is possible. The data was already yours. AI is the translator that finally lets you walk through the door.

PageMotor and Practical AI Updates

Free, informative, and FUN!

Frequently Asked Questions

What is the Musk-OpenAI trial actually about?

Musk committed $38 million to OpenAI when it was a nonprofit. The lawsuit alleges he was deceived when OpenAI converted to a for-profit. He is asking the $134 billion in claimed wrongful gains be redirected to OpenAI’s charitable arm, not paid to him personally. Read more below.

What changed in OpenAI’s deal with Microsoft?

On April 27, OpenAI and Microsoft jointly announced that Microsoft’s IP license is now non-exclusive through 2032. Azure is no longer the only cloud OpenAI runs on. Microsoft kept its ~27% equity stake but lost the cloud lock-in. Read more below.

What is GPT-5.5 (codename Spud)?

A frontier model OpenAI shipped on April 23, 2026. It scores 82.7% on Terminal-Bench 2.0 and ties or leads Claude Opus 4.7 and Gemini on several agentic benchmarks. Pro tier API pricing doubled. Read more below.

How did a Discord group get into Anthropic’s Mythos model?

Not through a clever jailbreak. The group guessed the URL based on Mercor breach patterns and authenticated using a third-party vendor contractor’s credentials. The danger was the perimeter, not the model. Read more below.

Can a non-developer actually use an API?

Yes, with AI as the translator. Olga walked through her real workflow: paste the API documentation link into Claude Code, give it three credentials, and ask it to write the test script. Zero code. Full setup in one evening. Read more below.

How much money went into AI funding this week?

$24.18 billion across 86 companies—the third-biggest week of 2026 and 14 times last week’s total. AI captured 82.2% of all worldwide venture funding, the highest share recorded across 22 weeks of tracking. Read more below.

Practical AI: I Thought APIs Were For Developers. I Was Wrong.

Key Definitions

What is an API?

An Application Programming Interface. A defined way for software to ask another piece of software for data or actions, without going through a graphical interface. APIs have existed since 1993. AI is the translator that lets non-developers use them in plain English.

What is MCP?

Model Context Protocol. An open standard from Anthropic that defines what commands an AI agent can use to talk to a connected application. If an API is the data pipe, MCP is the rulebook for what the agent is allowed to ask through it.

What is Agentic AI?

AI that does not just answer—it takes actions. Reads documentation, runs scripts, calls APIs, and executes a workflow without a human pressing each button. Salesforce reported employee-agent interaction compounding at 65% a month.

What is “Generative UI”?

The idea that AI generates a custom interface for you on the fly. Chris’s argument on this episode: that framing is the old world projecting itself onto the new one. The actual future is no UI at all. AI talks directly to data through an API and returns the answer you wanted.

Quotable Moments

AI doesn’t need a UI. It’s just data getting exchanged back and forth. The whole point of a UI is for you to tell the system some structured data. AI can just generate that data without needing a UI to have to click buttons to do it.

— Chris Pearson on the opening hook

If you read “Musk wants $134 billion” you got it wrong. He wants $134 billion redirected to charity. Different story.

— Olga Pechnenko on the Musk-OpenAI trial

I always was kind of sad that I wasn’t a technology person. This week playing with APIs, I felt like I was invited to this VIP party that previously I wasn’t invited to. And now I can go to any VIP party I want because I have Claude.

— Olga Pechnenko closing the deep dive

00:00 The Future of Software Has No UI

The episode opened with a thesis from Chris that set the spine for everything that followed. The world has been talking about “generative UI"—the idea that AI generates a custom interface on demand. Chris pushed back: that framing is the old world projecting itself onto the new one. The UI layer existed because humans needed a way to give software structured data. AI can produce that data directly. The actual future is no UI.

Key Takeaway: Every UI is a translator between human thought and structured data. AI removes the need for translation. That is why operators are about to do work that used to require a developer.

02:52 Musk-OpenAI Trial: What the $134B Figure Actually Means

The trial opened on April 28. Musk was the first witness and was on the stand for three days. Sam Altman, Greg Brockman, and Satya Nadella are all still scheduled to appear. Proceedings are not televised or photographed. The trial is expected to run roughly three weeks with a verdict possible by mid-May.

The $134 Billion Correction

Most newsletters reported that Musk is asking for $134 billion. The actual filing language asks that $134 billion in claimed wrongful gains be redirected to OpenAI’s charitable arm—not paid to Musk personally. Musk’s original investment was $38 million on the basis of OpenAI being a nonprofit. Reuters coverage.

Chris’s read on the verdict: OpenAI will IPO regardless. Too many companies have too much at stake. The real precedent the case sets is for every other nonprofit-converted-to-for-profit AI shop watching from the sidelines.

09:08 Three Rewrites in Seven Days: TBPN, Microsoft, Google

While Sam Altman was on the witness stand, his company quietly rewrote three of the biggest deals in AI history. Two of them this week. One of them three weeks ago that nobody connected to the rest until now.

The OpenAI-Microsoft Deal Rewrite

On April 27, 2026, Microsoft and OpenAI jointly announced that Microsoft’s IP license is now non-exclusive through 2032. Azure exclusivity is over. OpenAI gets multi-cloud access, explicitly including AWS Bedrock. Microsoft retained its ~27% equity stake but lost the cloud lock-in that had been forcing enterprise customers to migrate data to Azure to use OpenAI products.

The Anthropic side of the cap table got busier the same week. Google announced a $10 billion cash investment with another $30 billion contingent on milestones, plus 5 gigawatts of compute starting in 2027 via Google Cloud and Broadcom. Anthropic now has Amazon and Google on the cap table. Microsoft is not.

The third rewrite was older. Three weeks before the show, OpenAI quietly acquired TBPN (Technology Business Programming Network), a daily Silicon Valley livestream, for a reported figure in the low hundreds of millions. The pattern only became clear this week: Sam rewrote the cloud story (Microsoft), the founder story (the trial), and the media story (TBPN). Three boxes. Same seven days.

13:43 Sergey Brin Returns to Google

Brin came back to lead a DeepMind strike team focused on closing the coding gap with Anthropic. The internal directive: every Gemini engineer must use Google’s internal agents. The coding-tool race is now four-front: Claude Code, OpenAI Codex, Cursor, and Gemini.

15:35 GPT-5.5 Ships Under the Codename Spud

OpenAI shipped a brand-new frontier model on April 23 under the internal codename “Spud.” The rest of the world figured out it was GPT-5.5.

GPT-5.5 Benchmarks and Pricing

Terminal-Bench 2.0: 82.7%. Expert-SWE internal: 73.1%. Leads or ties Claude Opus 4.7 and Gemini on several agentic benchmarks. Pro tier API pricing doubled. Power users running production tests reported the model over-thinking simple prompts, with one user reporting 17 minutes of internal reasoning before a basic answer. Launch announcement.

Chris’s beat on this is the show’s recurring frontier-versus-practical theme. Benchmarks are climbing. The actual jobs people do—running a dashboard, completing a document—do not need 17 minutes of internal reasoning. Reasoning weights are eating procedural work.

17:38 The Mythos Leak: URL Guessing + a Contractor’s Credentials

Anthropic gated Mythos on Episode 36 because the company said it was too dangerous to release. This week the access path leaked.

How the leak actually happened: Not a clever jailbreak. The Discord group guessed the URL based on Mercor breach patterns, then authenticated using a third-party vendor contractor’s leaked credentials. They used the access for benign tasks (building simple websites) to avoid detection. Anthropic statement: “investigating… no evidence of impact on core systems.”

Chris’s read: this is a PR play above all. Employees have had access to Mythos since February 26 and nothing visible has changed. The story for operators is the perimeter, not the model—third-party contractors with API keys are now the soft side of every secure system.

19:25 China Robotics: 8-Hour Shifts at 100%

While American AI companies traded headlines, China shipped. New national standards under Beijing’s 15th Five-Year Plan explicitly prioritize embodied intelligence for supply-chain dominance. Chinese embodied robots completed full 8-hour tablet assembly shifts at 100% success rate inside one factory.

China Embodied AI in Q1 2026

$30 billion in disclosed embodied AI funding across 300+ rounds. 9 new $10B+ Chinese AI unicorns. China now ships more humanoid robots than the United States despite lower per-company valuations.

Chris’s frame: robotics is procedural, not philosophical. China has practical reasons to solve practical problems. American AI investment chases reasoning benchmarks while Chinese investment chases shifts on factory floors.

22:39 DeepSeek V4 + the $20B Raise

DeepSeek released its V4 preview on April 24 with a 1M context window at drastically lower cost than frontier U.S. models. The funding side told the bigger story: DeepSeek is in first external fundraising talks with Tencent and Alibaba at a reported $20 billion+ valuation. Same week, Mark Zuckerberg announced a $500 million CZI/Biohub partnership on AI biology with NVIDIA, Allen Institute, Broad, and Wellcome Sanger as partners.

27:14 Deep Dive: APIs Were Never Just for Developers

The deep dive opened with a confession. Olga has been running her recruiting firm for 13 years. Until this month she assumed APIs were a developer-only thing. Then she discovered APIs have existed since 1993. Salesforce launched its public web API in 2000. Stripe was API-first from day one. Twilio, Plaid, Algolia, MongoDB, Snowflake—all built API-first companies into multi-billion-dollar valuations while the rest of the SaaS world built UIs.

The Adoption Numbers

According to Cloudflare, ChatGPT crawler requests grew ~3,000% from May 2024 to May 2025. Perplexity bot requests grew ~157,000% in the same window. Salesforce reported employee-to-agent interaction growing 65% month-over-month and customer service via agent up over 2,000% year-over-year.

The framing Olga landed on: AI did not invent the unlock. AI is the translator. The data was always yours. The door was always there. Now you can finally walk through it without hiring a developer to read the manual for you.

38:44 The Frankenstein Dashboard

Before the API key, Olga’s team ran on a Google Sheet she calls the Frankenstein. Every recruiter manually entered numbers each morning for the day before. The team adds about 300 new candidates a week. About a month before the show she signed onto Loxo, a recruiting CRM that markets itself as AI-native. The reports inside Loxo were beautifully designed but isolated—each on its own page, each requiring a click and a context switch. Great UI. Wrong shape for an operator who wants to see the full funnel in one place.

Common mistake: Assuming a pretty UI removes friction. A great UI is still a UI. You still have to know what to click, where, and in what order. If your job is to coach a team, that overhead compounds every morning.

44:51 One Evening with Claude: Loxo API to Live Dashboard

While setting up reports inside Loxo, Olga asked her rep about API access. The rep handed over the credentials. Olga pasted the documentation link into Claude Code with one instruction: read this and tell me what you need. Claude asked for three things and explained where to find each one inside Loxo settings: the bearer token, the domain, and the agency slug. Claude wrote a 30-second test script. HTTP 200 came back. From there, Claude drove. It read the API documentation, mapped what was possible, tested writes by renaming a job, and documented the capability surface.

Pro tip: Olga used Claude Code in the terminal (CLI), but she verified the same workflow runs end-to-end in Claude Code Desktop. The CLI is faster once you are used to it. The desktop is friendlier on the way in.

The actual build took one evening and roughly five iterations on the visual design. Olga screenshot her existing Frankenstein and described what she wanted. Claude generated an HTML prototype. She reacted to the prototype—"I like this, I don’t like that"—instead of writing a spec. She drove the what. Claude did the how. The dashboard now refreshes every morning at 7 a.m. and lives in Dropbox so the team has live read access.

49:38 Red to Green in Three Days

The dashboard is color-coded by recruiter against a goal. Day one: most cells were red. Day two: less red. Day three: mostly green. Nothing in the dashboard changed across those three days. What changed was that Olga could see the gap, walked her team through the new workflow, and the recruiters started entering their sourcing data into Loxo instead of into the old sheet.

The Data Entry Reality

Industry estimates put unentered sales activity at roughly 74% of all sales work. The best AI tool on the planet cannot help if the data never enters the system. The dashboard did not fix the team. The dashboard made the gap visible enough that the team could fix it together.

The second view in the dashboard is the funnel: percentages, conversions, and benchmarks against goal, all on one page. Chris’s note on air—the “did you do this?” framing of the old sheet became “let’s look at this together"—shifted the dashboard from accusatory to investigative. Olga then ran a second report against the Loxo campaigns API: healthy campaigns versus campaigns with major data quality problems. The same data Loxo’s UI showed one campaign at a time, now in a single sortable view with patterns surfaced.

1:05:07 Bret Taylor, Linear, Stripe, PageMotor: API-First by Design

The pattern showing up across the industry: the people who built the SaaS UI lock-in are now building the way out. Bret Taylor, former co-CEO of Salesforce, is now CEO of Sierra, an AI-native customer service company explicitly designed to replace UI clicks. Sierra is at $150M ARR with 40% of the Fortune 500. Linear, the most beloved B2B UI of the decade, is also one of the most API-accessible products on the market. Stripe (the Collison brothers) built API-first payments into a $95 billion company by skipping the dashboard the rest of the industry built. PageMotor, the AI-native CMS Chris is building, applies the same pattern at the website layer—two websites, one for humans, one for AI agents.

Key Takeaway: The web hosting industry was built around WordPress for 25 years. Most legacy hosts cannot give an AI root-level access to a server. Only VPS-style hosts can. That means the next decade of websites moves to a current minority of hosts. The whole industry is about to change.

1:08:52 MCP Explained

Anthropic shipped MCP (Model Context Protocol) as an open standard. Chris’s analogy on air landed clean: if the API is the kitchen, MCP is the menu at the restaurant. It defines what the AI agent is allowed to order and the format the order needs to be in. If the SaaS tool already publishes good API documentation, MCP is not strictly required. But once the broader market standardizes on MCP, any agent can talk to any compliant application without the SaaS company building anything new.

1:18:11 Funding: $24.18B, 82% AI, 86 Companies

After five quiet weeks, the funding spike was back. $24.18 billion across 86 companies this week, 14 times last week’s total. Only one week in the 22-week tracker on this show was bigger, and that was OpenAI’s mega-round in March. AI captured 82.2% of all worldwide venture funding—the highest share recorded on the tracker. United States took 90% of AI rounds with 44 companies. Europe was second. China was third.

Top 5 AI Rounds This Week

Project Prometheus: $10B (stealth, AI for the physical economy). Anthropic: $10B (Google’s first tranche of the $40B commitment). Ineffable Intelligence: $1.1B seed in London—the largest AI seed of the entire 22-week tracker. True Anomaly: $600M (space security and defense AI). Robot Era: $200M (China embodied intelligence).

The pattern: physical AI is in a dominant lane. 7 of the top 16 rounds went to robots, drones, defense, and embodied AI. The Ep31 prediction—"AI gets hands"—has hit harder than the show called it at the time.

1:23:51 Big Takeaway and Prediction

Chris closed on the structural shift. The UI layer existed because humans needed a way to give software structured data. AI removes that requirement. Software expertise as a category is now closer to a video-game hobby than a professional credential.

The era of the software wizard is over. AI is the wizard now. You bring the outcome.

Chris’s prediction: humans will not enter data directly into software much longer. Data entry will run through an AI intermediary that ensures the data lands in the right column with the right format and pushes back when something is missing. The data integrity problem and the data entry problem solve together.

Olga’s takeaway: she always thought of herself as not-a-tech-person because she has a marketing degree. This week with APIs felt like getting invited to a VIP party that had been off-limits her whole career. Now she can walk into any of them. Pick one tool you already pay for. Ask if it has an API. Find out what your data has been waiting to tell you.

Keep Learning