How To Launch An AI Product Without Building An App (Use Slack) | Practical AI Ep 47

Published: June 26, 2026 · Hosts: Olga Pechnenko and Chris Pearson

Table of Contents

• About This Show
• Frequently Asked Questions
• Key Definitions
• Quotable Moments
• Where Will AI Actually Meet Normal People?
• What You Get From Practical AI Every Week
• The News: Model Loyalty Is Dead
• Sakana’s Fugu: One AI That Picks The Best Model
• OpenAI Says Non-Developers Now Use Codex 137x More
• Why Apple Just Raised Mac And iPad Prices
• $500M To Retrain Workers Before AI Hits
• Claude Becomes A Coworker Inside Slack
• Quick Hits: GPT-5.5, Gemini Notebooks And Computer Use
• Can You Trust AI To Check Its Own Work?
• Build Your Own AI Review Panel (Steal This)
• Slack Is Becoming The Launch Surface For AI Products
• Skip Building An App: Meet Customers Where They Already Are
• Salesforce vs Microsoft: Who Wins The AI Workspace
• The Catch: Own Your Brain, Use Slack As The Doorway
• Demo: What An AI Marketing Brain Reveals About Your Website
• Why Your Site Gets Zero AI Citations (And The Fix)
• Turning The Top Fix Into A Live Page In Seconds
• Funding: The Money Went To The AI Plumbing
• The Drop Was An Illusion: US AI Funding Tripled
• The Top 5 Rounds: Where The Billions Went
• Big Takeaways: One Interface, Less Friction, Ship Today
• Keep Learning

---

What You’ll Gain

• A concrete model for launching an AI product without building an app. Use Slack as your interface layer, own the brain behind it, and meet your customers where they already are.
• A reusable AI review panel you can run on any public-facing work. Three roles, three blind sessions, one pass. What more than one flags is the real problem.
• The straight story on Sakana Fugu, Claude Tag, Apple prices, and RAISE US. Corrected where the raw audio got it wrong.
• A live demo of taking a tool’s number-one recommendation and shipping it instantly. From finding the fix to a live page in under a minute, using PageMotor MCP.
• Week 30 funding in plain English. Which companies got the billions, what they do, and what it says about where AI is headed.

Biggest Takeaway to Implement: You do not need to build your own app to launch an AI product. Open a fresh Slack workspace, set up your agent, and let customers talk to it where they already live. Own the data and the brain. Rent the interface.

---

Key Definitions

Quotable Moments

I love my AI. I don’t trust my AI.

— Olga Pechnenko, on why the blind review panel matters

Never let AI grade its own homework. Open a brand new chat. Paste only the finished work, not the whole conversation.

— Olga Pechnenko, in the Steal This segment

Own your data and your brain. Use Slack as a doorway, not your foundation.

— Olga Pechnenko, on the right way to build on Slack

You’re the medieval queen sitting in her chair. They’re waving palm fronds at you and feeding you grapes and you tell the subjects what you want. You just expressed what you wanted and the thing that you wanted happened.

— Chris Pearson, on the AI-native way to operate

Model loyalty is dead. You don’t rely on one model. The one who can optimize the routing is going to win.

— Olga Pechnenko, on the news theme of the week

---

0:00 Where Will AI Actually Meet Normal People?

The episode opened on a question the show first raised months ago: where is AI actually going to meet people who are not using CLI tools, not in the Claude UI, and not running code? They are out there. They are the majority. And this episode, the answer started coming into focus. It is not a new app. It is Slack, Teams, and the tools they already have open all day. That shift changes everything about how you should think about building or launching an AI product.

1:12 What You Get From Practical AI Every Week

Olga set the context for anyone new: this is where you come to make sense of the AI space every Friday. The micro-economics of AI, the big-stage moves, and how all of it trickles down to practical application. She was honest about the week: learning on the bleeding edge is not always fun. There are bumps. She is going to share the good, the bad, and the ugly of what it actually looks like to operate here.

2:13 The News: Model Loyalty Is Dead

A lighter week for headline model announcements. The models are so capable now that the news is less about “which one is smarter” and more about implementation: where do they go, how do they integrate, who uses them and for what. The theme that emerged across every story: the era of picking one model and being loyal to it is over. Olga runs OpenClaw on Codex, Athena on Claude. She is already not loyal. The show is already not loyal. That pattern is spreading everywhere.

2:50 Sakana’s Fugu: One AI That Picks The Best Model For The Job

Japan-based Sakana AI released Fugu, and the launch post became the most-bookmarked AI topic of the week on X with roughly 30,000 saves. The concept: one API that automatically routes each task to the best model from a swappable pool, instead of locking you to a single one. Sakana claims frontier-level results without building one giant model. It is a productized version of being multi-model, the same way the Practical AI show already operates.

Olga pushed back honestly: Cursor and Lovable already do this. Chris read the corporate angle accurately. Cost management and token limits are the real pain driving this interest, particularly for large companies running high-volume workloads. Worth watching, not ready to call it transformative yet.

7:56 OpenAI Says Non-Developers Now Use Codex 137x More

OpenAI published its own internal data showing Codex is now the primary AI tool across every department, with non-developer adoption rising 137 times since August 2025, outpacing developers. Work is reportedly shifting from quick chats to long delegated tasks lasting 30 minutes to 8+ hours. Important framing: this is OpenAI’s own internal claim, not an independent study. It is a bakery reporting on how good its own muffins are. The trend is directionally believable. The specific numbers are OpenAI’s.

Chris identified who is actually doing this: entrepreneurs. People who used to need a developer to ship something now sidestep that friction entirely. That is the real story in the 137x number.

10:44 Why Apple Just Raised Mac And iPad Prices

Reuters reported June 25 that Apple raised prices on Macs, iPads, HomePod, and Apple TV by roughly $100 to $300. Apple’s stated reason: memory and storage chip costs are skyrocketing because AI data centers are consuming those components at scale. The chip race that the show has tracked for a long time just landed on regular consumers. Note: the iPhone was not included in this round of increases.

Chris summed it up: if you bought a MacBook in 2021, it is now like owning a 2018 Toyota 4Runner. You can probably sell it for more than you paid. Chris said he wished he had just bought one before this wave hit.

12:14 $500M To Retrain Workers Before AI Hits

AP reports a new bipartisan nonprofit called RAISE US launched with more than $500 million to pilot state-level training and transition programs ahead of AI-driven job displacement. Backers include Amazon, Microsoft, Anthropic, and the OpenAI Foundation. Important sourcing note: this figure and these backers come from AP’s wire reporting, not from the companies’ own official announcements.

Olga and Chris were both skeptical about what “retraining at scale” actually looks like and how much of the funding goes toward consulting versus real worker impact. The show will keep watching. The headline tells you these companies feel the urgency to be seen doing something.

14:14 Claude Becomes A Coworker Inside Slack

Anthropic officially announced Claude Tag: you @-mention @Claude in any Slack channel and it works as a persistent coworker, reading threads, breaking down tasks, and using connected tools. It runs on Opus 4.8. Currently available in Enterprise and Team Slack plans as a beta. It replaces the old Claude-in-Slack app, with a 30-day migration window for existing users.

Olga flagged the privacy angle directly: Claude reading every Slack message means teams have to be intentional about what goes in those channels. HR implications are real. Chris called it the potential for “the HR bot in the chat.” For small, agile teams with clear focus, this is a powerful workflow tool. For organizations with existing Slack culture issues, it adds a new layer of accountability people did not ask for.

17:07 Quick Hits: GPT-5.5, Gemini Notebooks And Computer Use

Three moves in quick succession. OpenAI sharpened GPT-5.5 Instant, making the default model better at advice, planning, and research. Google dropped free Study Notebooks inside the Gemini app, a no-cost space to study and organize content. And Gemini Computer Use preview arrived: Gemini can now click around a browser and take actions for you. Gemini is last in the browser-agent race but moving. Chris joked it will probably just click on its own ads.

19:21 Can You Trust AI To Check Its Own Work?

The first deep dive came from Dharmesh Shah (HubSpot co-founder), whose June 24 newsletter posed the question: can AI grade its own homework? His short answer, and Olga’s: no. Prompting AI to summarize something, do its best work, and check for mistakes is like asking a cat not to drink the cream. Of course it will tell you the work is great. It has no distance from what it just produced.

The mechanism that makes this worse: a long running session cannot catch errors in its own output because it has all the context of how the output was built. A fresh session with no context, given only the finished work, will find what the original session cannot see. That is the starting point for the panel approach.

24:16 Build Your Own AI Review Panel (Steal This)

Olga ran three blind AI reviewers on the show’s news roundup for this episode. None of them knew how the content was made, who made it, or what the others found. Three blank sessions, three different roles. The everyday viewer (busy professional, no jargon, tunes out fast). The tough producer (one thing will keep people watching or lose them). The relentless fact-checker (every label and number is a target).

What they caught in one pass: the lead story was being narrated like a spec sheet instead of grabbed like a hook. And the RAISE US source was labeled OFFICIAL when it is an AP wire, not a company announcement. That second catch was a real error, and the blind panel found it before the show aired.

30:29 Slack Is Becoming The Launch Surface For AI Products

The bigger pattern behind Claude Tag and Boring Agent is worth naming plainly: Slack is becoming the place where AI products get launched. Instead of building a full web app with its own login and onboarding flow, AI product makers are dropping their agent into Slack and letting teams talk to it where they already live. The Slackbot is quietly being rebuilt into an orchestration layer that routes between specialist agents.

Chris framed it clearly: imagine your company’s AI agent in Slack connected to PageMotor, to your CRM, and to your inventory. Someone sends a message. The agent handles the whole workflow end to end. No UI switching. No tickets. No “I sent it to the right person” delays. Just express the need and the need is met.

36:16 Skip Building An App: Meet Customers Where They Already Are

The core opportunity in one line: skip building a web app and reach customers inside the tools they already open every day. Every new app is a new login to forget, a new habit to form, a new reason to churn. Slack has 38 to 47 million users. Teams has 320 million. These are not platforms you need to grow. They are already the workplace.

The pattern that is emerging from summer 2026: agents are moving out of isolated browser tabs and into the shared context of where a team already works. Distribution shifts from pull (get them to come to you) to push (be there when they need you). For B2B AI products, this is a structural shift in how you reach your customer.

38:07 Salesforce vs Microsoft: Who Wins The AI Workspace

Salesforce bought Slack, and the integration is getting real. Slack is now provisioned on day one for every new Salesforce org. Slack’s own data (take with appropriate salt): AI-enabled Slack apps up 690% year-over-year, 1.7 million apps used weekly, 95% of users say apps make tools more valuable. Companies that build successfully inside Slack have historically been acquired by Salesforce, which is its own kind of growth strategy.

Olga’s bet: Salesforce is more AI-native than Microsoft in how it is building out this ecosystem, and Slack has the startup-friendly positioning that Teams does not. Teams is the 320-million-user incumbent for enterprise. Slack is the sharper, faster surface for the companies actually innovating right now.

44:50 The Catch: Own Your Brain, Use Slack As The Doorway

Slack banned training LLMs on Slack data in May 2025. Slack ships its own agents, which can compete with and squeeze yours. If your remote AI connection hiccups, your agent looks broken to every customer in that channel. And the platform controls your memory, your bill, and your reliability. These are the real risks of building on borrowed real estate.

The right frame, and the one James Dickerson (Boring Marketer) has already built into his product: the real product is the brain and the dashboard. Slack is just the doorway customers walk through to talk to it. Own your data and your intelligence. Use Slack as a coordination layer. That is the architecture that survives platform shifts.

47:02 Demo: What An AI Marketing Brain Reveals About Your Website

James Dickerson (@boringmarketer, boringmarketing.com) launched the Boring Agent: a credit-based AI marketing brain at $99 per 1,000 credits. You add it to Slack, point it at your site, and the agent does a full crawl of your marketing situation: search profile, LLM visibility, competitor gaps, and a ranked queue of opportunities. The real product is the web dashboard. Slack is where you interact with it and ask follow-up questions.

Olga ran PageMotor through it live. Key findings: 15 million monthly searches mapped in the space, PageMotor ranks for none of them, and zero AI citations across all the major AI engines. The dashboard also found 28 fixable website issues and a competitor gap in the agent SEO category. The Slack connection had a hiccup during the demo. That is the honest reality of shipping new products. The data was still rich and immediately actionable.

Chris’s read on why this product matters: SEO and marketing have been converging for 20 years. This tool acknowledges that they are now the same thing and builds accordingly. It gives you a discipline that lives separately from your CMS, feeds findings to your AI, and lets the AI act on them. That is the ecosystem that replaces the old SEO plugin.

52:40 Why Your Site Gets Zero AI Citations (And The Fix)

The Boring Agent’s analysis of PageMotor surfaced something important: the public site still describes the 2025 version of the product. Three months out of date in AI time is very out of date. AI engines read what is on the page. If the page describes an old version of your product, AI confidently tells people the old story. The fix is not more SEO tricks. It is making the front door current.

The ranked opportunity queue from the agent put one move at the top: own the agent SEO category. A full written brief came with it, specific enough to hand directly to an AI and execute. The lesson: a tool that surfaces ranked, actionable opportunities is worth more than a tool that gives you vague scores. Know the specific thing to fix, in the specific order to fix it.

1:03:38 Turning The Top Fix Into A Live Page In Seconds

Olga took the agent’s number-one recommendation, the agent SEO opportunity brief, and handed it to Claude via PageMotor’s MCP. Claude wrote the page. PageMotor published it. The page went live at pagemotor.com/agent-seo while the show was still running. From finding to live page: under a minute of her time. That is what zero friction between you and the outcome actually looks like when your AI is connected to your publishing layer.

Chris connected the dots: the Boring Agent produces a markdown file of directives, your AI reads the directives, PageMotor executes them. You could put this on a weekly cron job. Monday morning, the site has been updated based on what the agent found on Sunday. That is where this is heading.

1:10:55 Funding: The Money Went To The AI Plumbing

Week 30 of the tracker: approximately $4.68 billion went to AI across 77 companies, roughly 41% of all venture dollars. The theme this week was infrastructure. The money went to the companies that run AI models for others, not the companies that train them. As Chris explained: a lot of companies want to run their own AI but do not want to manage GPUs, data centers, or hardware deals. The companies that abstract that away are worth billions right now.

1:12:44 The Drop Was An Illusion: US AI Funding Tripled

The headline number looked like a crash compared to last week’s $11.64 billion. It was not. Last week was a China week, dominated by DeepSeek’s massive round. Strip that geography illusion out and US AI funding went from $1.3 billion to $3.7 billion, roughly tripling week over week. The US took about 80% of all AI dollars this week. The Middle East (essentially Israel) took most of the rest. China and Europe had quiet weeks.

1:13:09 The Top 5 Rounds: Where The Billions Went

The five largest rounds confirmed the infrastructure theme:

• Baseten. $1.5B Series F (San Francisco). Inference infrastructure: the plumbing that runs AI models inside other companies’ products, fast and in real time, without those companies having to build their own data center stack. The biggest check of the week went to running models, not training them.
• Groq. $650M (San Jose). Designed its own chips to run AI models fast and cheap. An alternative to Nvidia GPUs. Signed a deal with Nvidia the same week, which made the competitive picture complicated.
• Libai. $300M (Beijing). AI image generation and sharing platform. China continues to concentrate its AI investment in image, video, and robotics.
• Dream. $260M (Tel Aviv). AI cyber defense: helps governments protect critical infrastructure, power grids, and national systems. All three of the Middle East rounds this week were Israeli companies.
• Mirendil. $200M SEED (US). Ex-Anthropic and Google researchers building AI tools that help scientists create their own models. The anomaly of the week: a $200 million seed round valued at $1 billion. Seed rounds are getting absurd.

1:16:19 Big Takeaways: One Interface, Less Friction, Ship Today

Chris’s closing frame: think about your business workflows and what it would look like if an AI could be directed through whatever tool your team already uses. Maybe it is Slack. Maybe it is text messages. Maybe it is email. Every good company is eventually going to have one interface for their AI. You tell it what you need, it goes out to all the other services and handles it, and you get results. That is the direction. Start building toward it today.

Olga’s close: model loyalty is dead. The way building works has changed. You do not have to build your own app. You can borrow Slack as the interface and ship something real. The chip race landed in your wallet this week via Apple prices. RAISE US confirmed that even the largest companies feel the urgency of AI’s impact on workers. And the Boring Agent showed what happens when you give someone who has done the hard work access to your data. Act on the information you have. Ship the thing. Do not wait for the Slack connection to be perfect.

Note: Practical AI is taking one week off for the 4th of July. New episodes return in two weeks. Subscribe so you do not miss it.

---

Keep Learning

• Subscribe to Practical AI on YouTube. New episodes every Friday at 11am CT.
• The AI self-review panel (slides). Built from Dharmesh Shah’s subagent idea.
• Slides: Launch Your AI Product on Slack. The full Slack deep dive in deck form.
• AI Funding Report, Week 30. Every top-5 round verified against primary sources.
• The 30-Week AI Funding Tracker. The running total since the tracker launched.
• The agent-SEO page Olga shipped live on air. Boring Agent’s number-one recommendation, executed via PageMotor MCP.
• Sakana Fugu. The AI orchestration router that was the most-bookmarked AI post of the week.
• OpenAI on how agents are transforming work. The source for the 137x non-developer adoption claim.
• Anthropic: Introducing Claude Tag. The official announcement for Claude in Slack.
• Reuters: Apple raises Mac and iPad prices. Reuters quoting Apple’s official statement on the chip cost increase.
• AP: RAISE US $500M worker retraining fund. The AP wire report on the bipartisan nonprofit.
• Dharmesh Shah on X. The newsletter inspiration for the blind review panel deep dive.
• Boring Marketer (James Dickerson). The Boring Agent and marketing skills behind the demo.
• PageMotor. The AI-native CMS Olga and Chris build on, designed for talk-to-publish and the GEO era.