AI vs Software Subscription Mistake

Every time I call an Uber I'm reminded of AI's biggest lie. That $100 ride used to cost $3 in 2015. We got hooked on convenience while VCs subsidized our addiction. Now I'm watching the exact same playbook unfold with AI tools and most people have no idea what's coming. I keep thinking about a conversation I had with a startup founder last week. She was bragging about how her team of 3 was now doing the work of 15 people thanks to AI tools. "We're saving $800K in salaries," she said. "It's incredible." But here's what she didn't realize: She's living in the calm before the storm. 𝗪𝗲'𝗿𝗲 𝗶𝗻 𝘁𝗵𝗲 𝗨𝗯𝗲𝗿 𝟮𝟬𝟭𝟱 𝗺𝗼𝗺𝗲𝗻𝘁 𝗼𝗳 𝗔𝗜. Remember when Uber rides cost $3 across town? When they threw promo codes at us like confetti? Venture capital was bleeding money to get us addicted to convenience. Then the subsidies stopped. That $3 ride became $25, then $100 . We were hooked, so we paid. AI is following the exact same playbook, and the signs are everywhere: → OpenAI just hired a "CEO of Applications" (hello, monetization strategy) → Claude, ChatGPT, and others are still pricing at consumer rates despite enterprise-level capability → VCs have pumped $50+ billion into AI companies that need to show returns 𝗧𝗵𝗲 𝗿𝗲𝗮𝗹𝗶𝘁𝘆 𝗰𝗵𝗲𝗰𝗸 𝗶𝘀 𝗰𝗼𝗺𝗶𝗻𝗴 𝗳𝗮𝘀𝘁. When an AI tool can genuinely replace 2 full-time employees, it won't cost $20/month forever. It'll cost what those employees cost potentially $100K+ annually. Think I'm being dramatic? Look at enterprise software pricing. Salesforce charges $300/user/month. Adobe Creative Suite went from $50/month to $600/year per license. These companies price based on value delivered, not development costs. 𝗛𝗲𝗿𝗲'𝘀 𝘄𝗵𝗮𝘁 𝘀𝗺𝗮𝗿𝘁 𝗽𝗿𝗼𝗳𝗲𝘀𝘀𝗶𝗼𝗻𝗮𝗹𝘀 𝗮𝗿𝗲 𝗱𝗼𝗶𝗻𝗴 𝗿𝗶𝗴𝗵𝘁 𝗻𝗼𝘄: 𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝘄𝗵𝗶𝗹𝗲 𝗶𝘁'𝘀 𝗰𝗵𝗲𝗮𝗽. That side project you've been planning? That startup idea collecting dust? This is your window. AI development costs will never be this low again. 𝗦𝗵𝗮𝗿𝗽𝗲𝗻𝗶𝗻𝗴 𝗰𝗼𝗿𝗲 𝘀𝗸𝗶𝗹𝗹𝘀. When AI becomes prohibitively expensive for daily tasks, professionals who can write, analyze, code, and strategize without $1,000/month in AI subscriptions will command premium rates. 𝗖𝗿𝗲𝗮𝘁𝗶𝗻𝗴 𝗔𝗜-𝗮𝘀𝘀𝗶𝘀𝘁𝗲𝗱 𝘄𝗼𝗿𝗸𝗳𝗹𝗼𝘄𝘀 𝘁𝗵𝗮𝘁 𝗱𝗼𝗻'𝘁 𝗱𝗲𝗽𝗲𝗻𝗱 𝗼𝗻 𝗔𝗜. Learn the processes. Understand the thinking. Use AI to accelerate, not replace, your capabilities. The gold rush pricing won't last. The question is: Are you building wealth during the gold rush, or just getting addicted to cheap gold?

Replit's gross margins went from 36% to negative 14% in two months. Same product. Same pricing. Same team. The only thing that changed: they launched a more autonomous AI agent that consumed more LLM resources than their pricing covered. Traditional SaaS has 70-80% gross margins because one more subscriber costs almost nothing. AI products pay for compute on every prompt. Your best users are your most expensive users. That single fact breaks every pricing model designed for the SaaS era. I mapped pricing across the top 50 AI startups by valuation with Moe Ali. Six patterns emerged. The scariest finding: in most AI products, the P90 user costs 10-40x more than the P50 user. Both pay the same subscription. You're subsidizing your heaviest users with revenue from your lightest ones. And that subsidy grows as power users discover more ways to use the product. Cursor learned this the hard way. They switched from flat 500 requests/month to a credit pool system. A developer burned the entire monthly allocation in a single day. $7,225 invoice. The CEO published a public apology on July 4th. The plan description quietly changed from "Unlimited" to "Extended" twelve days after launch. Anthropic took a different approach. Their $17/$100/$200 tiers map to genuinely different user personas. A casual user, a power user, and a developer replacing an IDE. Those are different products with different willingness to pay. Then weekly rate limits targeting less than 5% of subscribers to push the heaviest users toward the API, where per-token pricing covers actual compute. The pattern across all 50 companies: pure flat pricing is dying. Nearly half use two or three models simultaneously. Here's the full breakdown: 1. Complete AI pricing guide: https://lnkd.in/gdKaQSMk 2. Replit guide: https://lnkd.in/gmA_c_AG 3. AI product strategy: https://lnkd.in/egemMhMF 4. AI agents guide for PMs: https://lnkd.in/eeey5Cxr If you can't estimate your cost distribution across P10 to P90, you're not ready to set a price.

Be careful what AI tools you sign up for… I learned this lesson the hard way so you don’t have to. When I first started experimenting with AI, I was excited. New tools popping up daily. Promising features. “Free trials.” Slick marketing. But here’s what I discovered after signing up for dozens of them ❌ Hidden usage fees that doubled my monthly bill ❌ Free plans that were basically unusable ❌ Locked-in workflows I couldn’t export ❌ Surprise charges buried in the fine print The hype can blind you. And I don’t want you to waste money (or time) like I did. Here’s what I look for now: ✅ Transparent, predictable pricing ✅ Real free plans with usable limits ✅ Data portability (export your workflows anytime) ✅ Clear documentation + active community AI tools can transform your business. But the wrong ones? They’ll drain your wallet before they ever save you time. Before you sign up for the next shiny AI tool, ask: “Is this tool saving me time, or just costing me more of it?”

I spent $100/month on Claude Code. Then $20 on Codex. Then Gemini. The models weren't the problem. The product boundary was. None of them let me switch to a cheaper model mid-session. Or route private prompts to a local model. Or fall back to another provider when one goes down. Each tool wants you inside its own garden, paying its price, using its model for every single task. That is backwards. So I built a model router into pi and measured 25 real sessions. 2,415 agent turns. 6 models across 3 providers. Total cost: $76.77. Sending every turn to GPT 5.5 would have cost $1,272.77. Kimi K2.6 handled 82% of turns for coding and reviews. DeepSeek covered logging and simple edits. Local Qwen handled private notes. GPT 5.5 only showed up when the cheaper path failed or the task needed deeper reasoning. The savings didn't come from one magic model. They came from not pretending every task is the same task. Tool-agnostic. Model-agnostic. Cheap by default. Expensive only when the task earns it. https://lnkd.in/g3Yd7gDd Where are you overspending on AI because your setup treats a linter and a system design session as identical?

Are you overpaying for AI tools? A Fresh FP&A client was and here's what we found...... They planned for $2.5K per month for AI tools. When we reviewed their financials, the reality was shocking. Their actual monthly spend was over $25K+ per month. The root cause was something every business owner and CFO needs to be aware of: Shadow AI. 😬 Employees were expensing individual subscriptions. Some teams were using multiple tools. Others were barely using any. API and token usage had wild variability. And costs were spread across what I call the New Big 4: Gemini Claude Microsoft Copilot ChatGPT All powerful tools but zero strategy and no oversight. Token costs are also rising because these models are getting smarter, processing more data, and handling more complex prompts. That means when your team is running long, multi-step prompts without visibility… Variable costs can BOOM. You cannot put AI subscriptions on a card and hope for the best. Here's how you take back control of your AI stack: 1️⃣ Centralize procurement and licensing 2️⃣ Monitor usage and set hard budget alerts 3️⃣ Establish clear governance policies 4️⃣ Train teams on efficient prompting 5️⃣ Develop a Business AI Roadmap AI tools are incredibly valuable. But without oversight, Shadow AI gets expensive. ❓Are actively monitoring AI spend, or is it a free-for-all right now?

External AI vendor: $400K a year. Internal AI build: $1.2M and counting. Guess which one is actually shipping in production. A CTO told me at dinner last quarter that they had spent eighteen months building an internal version of a voice AI tool they had been quoted $400K to license. "We figured we'd save money." Eighteen months in, $1.2M sunk cost, two senior engineers who had stopped wanting to work on it, and a system that broke every time a new model version dropped. He had just signed the vendor contract. This is the build-vs-buy math nobody puts on the slide. The license cost is what you negotiate. The maintenance cost is what your engineering team quietly absorbs until they stop absorbing it. Models change every quarter. Integrations change every release. The vendor's full-time job is keeping their system alive against that drift. Your team's full-time job is everything else. So the in-house build doesn't fail at v1. It fails at v3, when nobody on the original team is still on it and the new owner doesn't know which decisions were load-bearing. By the time the CTO calls the vendor back, they've lost a year of compounding value, and their team has lost confidence that AI works at all. The license fee is the cheapest part of buying. The maintenance bill is the most expensive part of building. Most companies invert that math, then pay both. InfiniTECX AI Marty Massih Sarim Vince T. Jack Madrid Nitesh Dubey Elaine Beeman Natalie Beckerman

The Cost of Intelligence in Business Software How should software companies charge for access to AI in their products? Should it be embedded in your per seat license fee (even if you pay more per license), or a separate cost based on usage, or maybe the value of the outputs? I don’t have the answers, but this is a growing issue among software companies who assume their customers will have fewer seats/employees as AI impacts jobs. And, it’s increasingly becoming an issue for business users who have to deal with unknowns around tech costs. On Ep. 193 of The Artificial Intelligence Show, I shared my personal frustration with the credit-based pricing model being used by companies such as HubSpot, Lovable, and Runway. We recently had to turn off AI features in HubSpot for a week to avoid going over some arbitrary credit limit that we weren’t even sure how we were hitting it. Now, to be clear, I love HubSpot, and have built multiple businesses with HubSpot as the core CRM. So, this is a wider industry issue that I’m illuminating through a personal experience. Here are a few of the challenges with credit-based pricing: Abstract and Variable The credits are abstract and variable, so we have no ability to forecast and manage them efficiently. For example, HubSpot’s credit pricing shows that the number of credits used varies based on the type of action being performed (e.g. customer agent interaction = 100 credits; data studio workflow with fewer than 500k rows = 25 credits; workflow automation = 10 credits). This becomes more variable in this model as the platform enables more reasoning, image generation, video generation, etc., which are more compute intensive than standard text and automation tasks. The Cost of Intelligence is Falling I know the cost of compute to serve up the intelligence in the product is plummeting (some estimates show an estimated annual cost reduction of 10x every 12 months). So if software companies are charging on a per token basis (what they pay through the API), will they continue to reduce the end user’s cost as their costs fall? That’s unlikely, which is annoying to know as a customer. Unknown Future Costs HubSpot has a section on their credit-pricing legal page showing a collection of beta AI features that they aren’t charging for yet, but will be soon. So basically every place we might use AI in HubSpot, we now have to try and consider how many tokens/credit might be used. Conclusion Again, I don’t know the answer, but generally speaking if a pricing model is causing more confusion than clarity, and makes me want to invest resources investigating alternative software solutions with more transparent pricing, it’s probably not ideal. I also feel like there is some element of the Bitter Lesson here in which software companies try to write a bunch of rules for how pricing should work (X credits for this, Y credits for that), when a more simple approach is likely the better solution for everyone. 

My team and I were burning money on our AI spend until I started actually tracking it. Now I check our balance regularly because I want to know exactly where those tokens are going, and make sure we're using them right, not just spending them. Most executives treat AI like software - a seat license with a fixed monthly cost. But AI isn't software. It's an operating cost. You pay by the token. Every sentence your AI writes runs up the bill, whether you keep it or throw it away. One unnamed company ran up $500 million in a single month on Claude because nobody set a spending limit. Uber burned through their entire 2026 AI coding budget in four months. Not edge cases. That's what happens when you manage a variable cost like a fixed one. So, here's what you should actually do about it: >> Track your waste rate: Generate ten versions of something and keep one? You paid for all ten. That's a 90% waste rate. At scale, that's serious money nobody booked. >> Right-size the model: Stop driving a semi to buy milk. Use Haiku for sorting and drafts. Use Opus for decisions where being wrong is costly. Match the tool to the job. >> Set guardrails before the bill sets them for you: A spending limit your AI enforces on itself. Not after the surprise invoice. Before it. Read my latest newsletter for the full breakdown on how to track, right-size, and govern your AI spend, with real examples of what works. 👇 #ArtificialIntelligence #Budgeting #productivity #AImarketing #ContentCreation

In SaaS, high usage is a win. In AI, it can bankrupt you. That is the AI pricing paradox. SaaS thrives on flat-fee pricing because marginal costs are near zero, but AI is different. Every interaction burns real compute, and costs scale with usage. If you’re not charging in a way that scales with usage, your best customers become your biggest cost centers. When Sam Altman revealed that OpenAI was losing money on ChatGPT Pro, even at $200/month, it was a wake-up call for every AI company. So what actually works? ∙ Usage-based pricing: Covers costs and protects margins. ∙ Credit-based or “tokenization” models: Simplifies usage metrics. ∙ Tiered access: Uses qualitative differences to capture more revenue. ∙ Hybrid approaches: Offers both predictability and scalability. The path to sustainable revenue relies on understanding the economics of AI and pricing correctly. Learn more about the AI pricing paradox and how to overcome it: https://lnkd.in/dEZa8n3A

Someone just canceled 5 AI subscriptions and hired 2 developers instead. They're calling it an "AI layoff." I love this industry. AI subscriptions are starting to feel like SaaS did in 2019. You add one for a good reason, then another, then another. Before you realize it you have five tools, five context windows that forget everything, five rate limits to babysit, and a combined bill that quietly crossed the cost of a junior hire. The promise was that AI would replace headcount. The reality for most teams is that AI created a new category of overhead nobody budgeted for. This does not mean AI is not worth it. For the right workflows it absolutely is. But the ROI calculation is more honest than the pitch deck version. AI works best when it is doing something specific, repeatable, and well-defined inside a system someone actually understands. When it is doing all of that, it is genuinely cheaper than a developer. When it is not, you end up paying for five subscriptions and hiring the developers anyway.