Two Product Management patterns that worked “Do Less” and “Validate a Lot”
Background:
You remember the disasters. Samsung phones catching fire. Cybertruck windows shattering on stage. Quibi burning $1.75 billion in eight months.
2025 had some of the sharpest product launches in years. Almost every one broke a rule that PMs treat as gospel.
I spent weeks analyzing what happened. The specific decisions that made each one work—and how the teams knew they were working.
The pattern: the 7 best products won by doing less.
But “do less” is dangerous advice without context. Here’s how seven companies got it right.
1. SURI Electric Toothbrush
Result $30M revenue in two years
Competition Oral-B and Philips own 75% of the market
Key Metric 20% of sales from word-of-mouth
The decision:
The founders (both ex-P&G) ran thousands of surveys asking one question: what do you hate about your current toothbrush?
The answers were specific:
- Batteries die too fast
- Travel cases are bulky
- Gross gunk builds up on the brush
- Can’t recycle the heads
SURI fixed all four. 40-day battery. UV-sanitizing travel case. Stainless steel plate that wipes clean. Plant-based recyclable heads.
No Bluetooth. No AI brushing coach. No gamification.
Jony Ive owns one. So do the Kardashians.
How they validated:
They tracked word-of-mouth as a percentage of sales. When it hit 20%, they knew product-market fit was real.
They also went to dentists for technical validation. One dentist told them a lighter brush reduces hand strain. Another suggested a softer mode for new electric toothbrush users. These insights never appeared in consumer surveys.
24 manufacturers said their plant-based brush head was impossible. Factory 24 said yes.
My learning:
Hatred is specific and actionable. Build a hate list. Tie every feature decision to it.
2. Google Pixel 10 Pro
Result “Best Android phone of 2025” from multiple reviewers
Competition Samsung Galaxy S25, iPhone 17
Key Metric AI suggestion engagement rate informed feature visibility
The decision:
The Pixel’s AI features are context-aware. Camera Coach gives composition tips while you’re framing a shot, then disappears. Magic Cue surfaces shortcuts based on current activity, then gets out of the way.
Google also stopped optimizing for benchmark scores. The Tensor G6 chip runs ML workloads efficiently. Synthetic tests don’t reflect real usage patterns anyway.
How they validated:
Google tracked when users engaged with AI suggestions versus dismissed them. This behavioral data trained the models to recognize when help was unwanted.
Reviewers praised the Pixel for AI that “didn’t get in the way.” That language appeared in zero Samsung reviews.
Selling it internally:
The Tensor chip team proposed optimizing for ML workloads over raw speed. Leadership accepted bad benchmark headlines on launch day in exchange for better experience reviews a month later.
My learning:
For every feature, define when it should disappear. Features without an exit condition become annoyances.
3. Cognition’s Devin AI
Result $1M to $73M ARR in nine months
Competition GitHub Copilot, Claude Code, Cursor
Key Metric 67% PR merge rate (up from 34% previous year)
The decision:
Devin’s positioning is specific: “An AI software developer that excels at tasks with clear requirements and verifiable outcomes that would take a junior engineer 4-8 hours.”
They targeted enterprises with massive legacy codebases. Goldman Sachs. Santander. Banks with millions of lines of code needing migration and security patches.
One customer saved 5-10% of developer time on security fixes. Another hit 20x efficiency on vulnerability remediation.
How they validated:
When independent testers found Devin succeeded only 15% of the time on complex tasks, Cognition published that number. They also published the 67% PR merge rate.
Enterprise sales cycles shortened. As one buyer said: “Every other vendor claimed magic. Cognition told us exactly when their product would fail. That’s who I trust with my codebase.”
Selling it internally:
The founding team went through eight pivots before Devin. They’d learned that overpromising kills companies.
Internal rule: never claim capabilities you can’t prove in a live demo. This meant walking away from deals where prospects wanted promises Devin couldn’t keep.
My learning:
Publish your failure rates. The companies that trust you with critical systems are the ones who’ve seen you be honest about limitations.
4. Perplexity AI
Result 780M monthly queries, $20B valuation, 38 employees
Competition Google Search, ChatGPT
Key Metric 640% YoY user growth in India through Airtel partnership
The decision:
CEO Aravind Srinivas defined Perplexity as an “answer engine.” You get synthesized answers with citations. You can verify claims.
Their growth strategy: distribution partnerships. They bundled Perplexity Pro with Airtel telecom subscriptions in India. Now they’re replicating with Samsung, Motorola, and possibly Apple.
Pre-installation beats app store competition.
How they validated:
Two metrics: queries per user (engagement depth) and query growth month-over-month.
Queries per user trended up every month. Users were replacing their Google habit.
The founding moment came from personal frustration. Srinivas was trying to find car insurance and couldn’t get a straight answer from Google. The team had a prototype. He tried it. It worked.
Seven days after ChatGPT launched, Perplexity launched publicly. They rode the wave.
Selling it internally:
Srinivas built a culture around shipping 80% perfect products. Each release becomes a foundation for improvement.
He ditched pitch decks. He showed investors the product directly. Let them use it. That authenticity converted skeptics.
The timing factor:
Dozens of AI search startups launched in the same window. Most are dead. Perplexity survived because of positioning and distribution, not just timing.
My learning:
Create new categories. Fighting incumbents on their terms is a losing game.
5. Cursor
Result $1B ARR by November 2025, $29.3B valuation
Competition GitHub Copilot, VS Code, JetBrains
Key Metric Zero marketing spend; 100% word-of-mouth growth
The decision:
Four MIT grads faced a choice: build a VS Code extension or fork VS Code entirely.
They forked. Extensions inherit architectural constraints. A fork gave them complete control over the experience.
Within three months of writing the first line of code, they shipped.
By February, every Coinbase engineer had used it. Stripe adoption went from hundreds to thousands. OpenAI, Midjourney, Perplexity, Shopify followed.
OpenAI tried to acquire them. Cursor said no.
How they validated:
They tracked one thing: were developers switching their primary editor to Cursor?
When engineers at Stripe and Coinbase made Cursor their default, the product was genuinely better.
Zero marketing spend was a signal, not a strategy. If growth required marketing, the product wasn’t good enough.
The failure they avoided:
Before Cursor, the same team spent a year building AI tools for mechanical engineering. 3D autocomplete for CAD systems.
They weren’t mechanical engineers. They couldn’t feel what was wrong because they didn’t live the problem.
Cursor worked because the founders were their own target users. They used Cursor to build Cursor.
My learning:
Forking makes sense when you’re building for yourself. If you’re building for users you don’t understand, the extension path lets you learn before you commit.
6. iPhone 17
Result#1 smartphone position, 247M units projected
Competition Samsung Galaxy S25, Google Pixel 10
Key Metric Base model sales nearly doubled in China vs. iPhone 16
The decision:
For years, Apple held back features to push people toward Pro models. 120Hz display, Always-On screen—Pro only.
The iPhone 17 broke that pattern. The base model got 120Hz, Always-On, 256GB storage, and the 48MP camera system. Same $799 price.
Apple’s data showed: base iPhone customers weren’t cross-shopping Pro. They were cross-shopping Android. Or waiting for discounts. Or not upgrading.
The differentiation strategy was losing customers, not upselling them.
How they validated:
When early sales data showed base iPhone 17 outperforming Pro models, Apple adjusted manufacturing from 25% base / 65% Pro to a more balanced split.
China was the proof. Base model sales nearly doubled. These were customers Apple had been losing to Huawei and Xiaomi.
Selling it internally:
Apple’s retail stores function as research labs. Store employees reported the same pattern: customers wanted Pro features but not Pro prices, and they were walking out.
When retail data, sales data, and China market data all aligned, the decision became inevitable.
My learning:
Feature gating works when customers upgrade. When they leave, you’re just losing. Check your data.
7. Nintendo Switch 2
Result 3.5M units in four days, fastest-selling Nintendo console ever
Competition PlayStation 5, Xbox Series X, Steam Deck
Key Metric Backward compatibility with 150M existing Switch owners’ libraries
The decision:
Nintendo fixed every complaint about the original Switch.
- Joy-Con drift? Solved with magnetic controllers.
- Small screen? 7.9 inches with 120Hz.
- 32GB storage? 256GB.
- Graphics gap? 4K output when docked.
Full backward compatibility with existing game libraries.
At $449, the Switch 2 costs 50% more than the original’s launch price. Nintendo bet that value would justify the premium.
How they validated:
A 2024 survey found 40% of Switch owners experienced Joy-Con drift. Nintendo’s president publicly apologized. Each complaint became a design requirement.
The magnetic Joy-Con almost happened in the original Switch. Prototypes went to then-President Satoru Iwata. The magnets were too weak. They scrapped the idea.
The team researched magnets for a decade. When the technology improved, they tried again.
The invite-only pre-order system generated data while generating sales. Which demographics signed up? How quickly did invites convert?
Post-launch surveys continue. The validation loop never ends.
When this approach fails:
Nintendo could refine because 150M people already loved the Switch. The formula was proven.
If your v1 hasn’t found product-market fit, “fix the complaints” might mean perfecting something nobody wants. Refinement is for products that work.
My learning:
Customers sometimes want the thing they love, but better. This only applies when they already love it.
When “Do Less” Fails
Four failure modes:
1. You don’t understand your users
Cursor’s founders could fork because they were developers building for developers. Their previous startup—AI for mechanical engineering—failed because they lacked that intuition.
Guardrail: Only apply restraint when you deeply understand the problem space.
2. Your v1 hasn’t found product-market fit
Nintendo could refine because the Switch was successful. 150M units. Clear feedback.
If you don’t have traction, “fix complaints” might mean perfecting something nobody wants.
Guardrail: Restraint is for products that work. Products that don’t need experimentation.
3. Timing is the actual variable
Perplexity launched seven days after ChatGPT. Dozens of competitors launched in the same window. Most failed.
Timing got attention. Strategy kept them alive.
Guardrail: If your market window is closing, speed beats polish.
4. Stakeholders confuse restraint with laziness
“We’re doing less” sounds like “we’re not trying.” Every company in this list tied decisions to specific data.
Guardrail: Document why you’re not building something. Restraint requires more justification than ambition.
How to Sell “Do Less” Internally
1. Tie every decision to customer evidence
SURI responded to every feature request: “Is it on the hate list?” Create your version.
2. Reframe restraint as focus
Google’s Pixel team: “We’re optimizing for real-world experience, which is what reviews measure.”
3. Set up fast feedback loops
Perplexity shipped 80% products because they could learn from real users within days.
If your feedback loops are slow, you need more upfront validation. Restraint is riskier when you can’t course-correct.
4. Show the cost of complexity
Every feature has maintenance cost, documentation cost, support cost, cognitive load cost. Most roadmap conversations ignore these.
Calculate the true cost of features you’re not building.
5. Find your internal champion
Someone with authority needs to believe in restraint. At Apple, leadership accepted bad benchmark headlines. At Cursor, the founders were the authority.
Five Questions for Monday
- What do your customers hate about the current solution? Build the hate list.
- For each feature: when should it disappear?
- Is your premium strategy converting customers or losing them?
- Can you publish your failure rates? If not, why are you promising things you can’t deliver?
- Is this a refinement problem or an innovation problem? Do you have product-market fit?
The Bottom Line
The pressure in product management is always to add features, add innovation, add complexity.
Your stakeholders want more. Your competitors are shipping more. The tech press rewards “breakthrough” and “revolutionary.”
The scoreboard cares about whether people buy.
The best launches of 2025 removed friction, restrained scope, and solved real problems.
Saying “no” requires more conviction than saying “yes.” Shipping less requires more justification than shipping more.
When you get it right, you build products people love and tell their friends about.
What’s the hardest “no” you’ve defended on a product decision? I’ll reply to the best answers.
If this changed how you think about product decisions, share it with your team. For deeper frameworks on validation and stakeholder management, check out my Advanced Certified Scrum Product Owner Course. Link in bio.
TL;DR
The 7 smartest launches of 2025:
- SURI — Built from customer hate list
- Google Pixel — AI that disappears when irrelevant
- Devin AI — Published failure rates openly
- Perplexity — Created “answer engine” category
- Cursor — Forked because founders were their own users
- iPhone 17 — Upgraded base model to stop losing customers
- Switch 2 — Refined proven formula over 10 years
Guardrails: Know your users deeply. Have product-market fit before refining. Tie every decision to data. Sell restraint as focus.
The question: What would you stop building if you had permission?
