Why Excel is not enough for AI SaaS ideas

Quote from Andrew Ng that nails what we’ve felt at Skarbe in 2025: “If you can build a prototype in a day but wait a week for product feedback, that’s your real bottleneck” This is exactly what happened to us in 2024/2025. With gen AI, we can ship features insanely fast. But what slowed us down, and still does, is deciding what’s worth building at all. We’re now at almost 300 customers. Every week, I get a flood of raw, unfiltered feedback. It’s constant. And honestly, that’s what’s shaping the product way more than anything we imagined at the whiteboard stage. My biggest lesson: -> Product owners are now the bottleneck, not engineers. -> You can ship features in hours. But figuring out what actually solves a real problem (and what to say “no” to) takes deep customer empathy and fast decision-making. Most days, that’s on me + Alex (co-foudner and CTO) I’ve spent 6 years as a PM and product leader in B2B saas unicorn, thought I “got it.” But nothing humbles you like hundreds of real users giving feedback, often pulling you in different directions. It’s not about collecting votes or shipping requests. It’s about feeling the customer’s pain and quickly turning that signal into a better product. What works for us now: -> Building tight feedback loops: weekly check-in calls, chat groups, emails, real voices, not just data. -> Giving the product owner real power to say no, focus, and act fast. Waiting for consensus = dead end. -> Shaping the roadmap in real time, not quarterly, not monthly, but actually IN REAL TIME! Bottom line: -> AI makes building fast. The real work is listening, deciding, and shaping. Every. Day. -> Product ownership isn’t just a job, it’s the constraint for every AI startup right now. Any folks out there feeling this shift as you scale?

AI business models are the final nail in the ⚰️ for ARR - so what should be the software industry's new flagship metric? Here's a simple alternative founders can use to impress at their next Board Meeting: Revenue Run-Rate Range (R3) = [Lower Bound, Midpoint Estimate, Upper Bound] Breaking this down into three metrics: - Lower Bound = sum of all contractual minimums, committed credits, and subscription baselines, annualized. A startup's ARR based on traditional formula. - Upper Bound = last 30-90 days of revenue, annualized. A startup's current Annualized Run-Rate Revenue (ARRR) - Midpoint = weighted estimate based on historical variance (stable history → lower variance → higher weight to upper bound) or cohort maturity (mature customers → more predictable). This measure is useful because it acknowledges the uncertainty of consumption- or outcome-based pricing models. The statement "We have an R3 of $14M-$22M-$28M" tells your Board much more than a single, falsely precise number. I also like that it helps with planning - a startup can budget with the lower bound as a worst-case scenario, confidently make investments based on the midpoint, and build capacity to handle the upper bound. There's been a lot of talk about the death of ARR from top startup minds like Dave Kellogg Kyle Poyar CJ Gustafson - curious to hear everyone's thoughts about R3 as a next evolution for measuring topline performance.

Most software is still built like it’s 2005 (great year btw). “Build 90% once, sell everywhere.” I’ll admit that playbook paid my bills for a long time. (It even helped me invest in venture capital while covering too many group bar tabs.) But in AI, that model keeps you stuck in endless POCs. From what I see, the teams that succeed treat the product as scaffolding to co-create with customers at scale. → Start with a core base = Faster adoption → Adapt it for each client’s specific use-case = Stickier retention → Align to the exact outcome they’re chasing = Competitors can’t copy it Fun part - it's typically worth 3-10x more in ARR :) Do you think this model scales, or does AI need something entirely different? Check out some examples of this being done precisely via TheAgentic Launchpad.

Faster tools, quicker timelines, instant results. Design has never been this fast. But is it still thoughtful? I see this tension everywhere.  Design sprints that prioritize shipping over solving. Founders that launch in weeks but rarely pause to ask: does this really matter? Execution today is frictionless with templates, AI tools, frameworks, no-code platforms. The “how” of building has never been easier, but the speed alone rarely creates something lasting.  •⁠ ⁠That design polished enough to impress but forgotten a week later. •⁠ ⁠⁠That product pushed out on time but empty of meaning.  •⁠ ⁠⁠That feature that shipped fast but solved nothing. The missing piece was the depth.  It comes only from asking "why" before the "how" It’s the decision to refine, not just release. It’s slowing down to understand the problem, the user, the feeling behind the click. As founders, it’s tempting to wear “speed” as a badge of honour. In the long run, people won’t remember how quickly you shipped. They’ll remember whether it worked for them 🙌

The documentation advantage that'll separate winners from losers in 2026. Everyone's building with AI and thinking they're ahead of the game. They're missing the real goldmine sitting right in their codebase. While your competitors are winging it with surface-level product knowledge, you're about to have an unfair advantage they can't even see coming. Here's what's happening Most startups work like this: → Developers build in isolation with AI tools → Product context lives in their heads and .cursor/rules files → Marketing team gets a brief and guesses at positioning → Sales team demos features without understanding the why → Everyone's working with different versions of "what we built" But here's the plot twist... All that structured context you're creating for AI coding? It's not just development documentation. It's your marketing strategy. Your sales playbook. Your competitive moat. Take your technical planning docs and feed them to AI: "Transform this product architecture into customer pain point stories" "Convert these feature decisions into positioning insights" "Turn these user flow docs into sales demo scripts" "Create a support team that knows your product inside and out" Suddenly your marketing team understands the product at a level your competitors can't touch. Your sales team speaks to real technical value instead of fluffy benefits. Your whole company operates from the same source of truth. While everyone else is playing telephone between departments, you're running a precision startup machine. The companies that figure this out in 2025 will dominate 2026. The ones that don't? They'll keep wondering why their product-market fit feels fuzzy while yours is laser-focused. Most founders think documentation is overhead. Smart founders realize it's their secret weapon. Are you building context that only helps your AI code better, or context that transforms your entire business? Want to start building context? Checkout the early access waitlist Precursor

⚠️ You can build an AI product in 48 hours now. That’s exactly why so many founders are screwing it up. With tools like Vibe Coding, shipping has never been easier. You can go from idea to live product in days - not months. But here’s the shift: ❌ The question isn’t can you build it. ✅ The question is: should you? I’m speaking with multiple founders right now who are skipping the customer development part entirely. They’re charging ahead, building features - without first checking if anyone actually wants them. No interviews. No validation. Just vibes and velocity. 🚀 Tools evolve. But principles don’t. Lean Startup thinking is more important than ever: Validate the problem before you write a single line of code. Talk to customers before you commit to features. Use prototypes as a learning tool, not a vanity milestone. The upside? You can get to something tangible faster than ever. Which means you can get to truth faster too. 💡 Lesson: The real advantage isn’t speed to product. It’s speed to validation. 📸 Pictured: Me with the author in London, 2012. 🕵️♂️ Bonus points if you can guess which book launch it was. Drop it in the comments 👇

“The first decision isn’t what to build. It’s who you want to help" Conor Brennan-Burke and his co-founder Manu Ebert initially set out to build Echo AI, an AI agent for product managers. They found the hardest part wasn't building the agent itself but ensuring it had the right context to provide value in every situation. After spending months developing a solution, they saw every founder around them struggling with the same problem. So they pivoted to build Hyperspell and help founders deploy agents on a stronger data foundation. Less than a year later, they're partnering with companies building on-screen assistants, embedded customer support bots, and CRM personalization tools, helping them save months of development. Learn more about Conor's founder journey and the invaluable lessons he's learned along the way in our latest founder blog. This is definitely an Every-backed founder story worth reading. Link in comments 💡

This is my exact process for building MVPs (in 60 sec) When you hear “build an MVP,” you probably imagine months of coding, endless features, and thousands of dollars. But in truth, the process can be broken into minutes—not months. Here’s what I everyone do (the hard way): - Spend weeks drafting business plans. - Hire developers before having users. - Add every feature I could think of. - Delay launching until it was “perfect.” - Burn money before learning anything. Here’s what I do now with AI Possibilities (the simple way): - Start with one clear problem to solve. - Sketch the idea on paper. - Build a wireframe. - Use Lovable or Replit for the first version. - Show it to 5 users immediately. The difference is staggering. One process is rooted in pride and fear. The other is rooted in clarity and speed. And only one actually creates traction. Simple beats complex when building MVPs. PS: If you had to launch in 60 seconds, where would you start?

Our Teams aren’t just building tools, they’re building living, breathing extensions of themselves that plug straight into the company’s Operating System. Same applies to our live Products like AlphaAI, which relies on a MCP repository enabling composable ecosystem tools, services, apps... Tobias Lütke dropped a real nice update on how they building with AI at Shopify. He's always on point, and what he shared reflects quite well the way we operate now. By now it should be clear to anyone building with AI that MCPs aren’t just a buzz word for API connectors, but that they’re the backbone of a new kind of Operating Systems. In our own experience, here’s what we’re living, not just spotting: Every team has been building MCP-powered tools, Creative Production Agent by the Art Team, the AlphaAI mesh on the Product side, our VibeOps Stack for accelerated internal development, you get the idea. Each tool/interface + associated MCPs are a direct extension of a team’s expertise, packaged as UX/UI and ready to plug into the Company, the Live Product mesh, or sometimes both!! What's been mindblowing to me is that Teams don’t just share data, they share the playbook. Through MCP enabled tools, each team gives access to the instructions, workflows, and know-how, so you work with their data like an insider. It’s like getting an expert in a box. Scout, shared by @tobi is a perfect example of that. LLM flows, knowledge bases, interfaces, databases, they're all compounding through MCPs. Not just linking, but multiplying what’s possible, which is CRAZY. Building this way is now intentional for us, not just accidental. The MCP repository has become the gateway to knowledge and capability, both inside and outside the company. And it's ran by experts in their fields, not replacing anyone. Is this the new standard? Absolutely. I will share more about what we're doing in the near future, there's just too much going on. Tobi's original post: https://lnkd.in/gCuaiibS

Everyone says “learn n8n.” I say most people shouldn’t have to. Automation shouldn’t feel like wiring a spaceship at 2 am. Webhooks. Nodes. Silent failures. This week, I tried something that flips the table. Google’s Opal. Describe the workflow in plain English. Watch a living canvas assemble it in front of you. Data pulsing through each step in real time. Prompt. Generate. Refine. Ship. I typed a simple intent: when a lead comes in, enrich it, notify Slack, create a task, send a follow-up. Opal built the skeleton in seconds. I dragged a few cards. Swapped one. Hit run. The canvas lit up like a neon grid. You can literally see the data breathing. Here’s what makes it different: - 🎛️ Intent-first building: you speak, it drafts the workflow. - 👀 Visual flow playback: watch each transformation as it happens. - 🧠 Google’s latest AI inside for text, video, and images. - 🧩 Remix-ready gallery: start from proven templates, then morph. - 🤝 Instant collaboration: share a link, co-build in real time. This is how builders win in 2025. Not by memorizing nodes. By shaping intent, then sculpting. Old paradigm: tool-first. Learn syntax. Fight the graph. New paradigm: goal-first. Explain. See. Edit. Ship. If automation were finally accessible to the entire team, what would you build first? A lead router that improves every week. A content pipeline from idea to publish. A support triage that learns your tone. This is bigger than a new tool. It’s a shift from node-first to intent-first work. Teams that adopt this mindset will compound, fast. I’m here for it. Are you? #AI #Automation #NoCode #GoogleAI #Gemini #Productivity #Workflow #BusinessStrategy #FutureOfWork #Builders