The press around AI putting pressure on well-established SaaS companies is gaining some momentum.

Note: We are not discussing specific stocks and valuations. Our focus is on the impact of AI on software companies.

Analyst Ratings Published 08/11/2025…

"Melius Research downgraded Adobe… warns of ongoing multiple compression for software-as-a-service companies… ‘AI is eating software’ …”

AI isn’t a shiny add-on anymore. It’s like a sneaky wave that’s pushing SaaS valuations lower. Investors see that and say: This might be more than a passing trend.

“Redburn-Atlantic downgraded shares to Sell, warning that generative AI tools are eroding the company’s competitive advantage …”

When analysts note genAI is eating away at your moat, that’s a red flag. For product folks, that means no more resting on past glory. You have to rethink the value your product brings.

On a related note:

“Atlassian laid off about 150 employees… AI will replace many support and operations roles to cut costs and boost efficiency.”

Jobs are vanishing because AI can now handle repetitive tasks.

---

For us on the product side, that points to two things: rising expectations for automation and shrinking tolerance for any friction.

These moves aren’t just headlines. They reflect a shift in how markets see SaaS. Using Adobe as an example, there are several software companies facing similar scrutiny.

When AI starts chiming that its tools can do your job faster or cheaper, investors get jittery. That changes how product managers should think: it’s no longer enough to polish features. AI-first competitors or internal automation can upend your roadmap in weeks.

The message from Wall Street? If the AI around your space is improving fast, your users could be just one model away from switching. And it’s not just about competition—it’s about cost, integration, and perceived value. If your AI isn’t sticky, it’s replaceable.

For product leaders reading this, here’s what’s buzzing in the market:

• Valuation risk is real, even for heavy hitters. If you don’t keep your competitive edge, investors will punish multiples fast.
• Expect AI to be a force multiplier or a replacement. Your value-props need to lean into what only you can do—be it deep domain context or specific workflows.
• Speed matters. Markets are reacting now, not in some distant “AI future.” You’ve got to move on to experiments, integrations, or whatever keeps your product relevant.

This isn’t about doom-scrolling or hype. It’s about reading the room. Adobe’s drop says investors are cautious. Atlassian’s restructuring signals that internal teams are already adapting to AI. By now, you should all be seeing these shifting trends in your product areas and business.

As product managers, those aren’t distant storms—they’re the new normal.

From WSJ piece on Warby Parker:

Many things have gotten pricier in the past 15 years. Not Warby Parker's most affordable glasses, which have cost $95 since the brand’s inception in 2010.

Warby Parker grew 14% last year. It did this while keeping its hero $95 price point. This shows that a focused value proposition can thrive even with inflation. The company used a few key strategies. It controlled its supply chain. It created a tiered pricing model. It constantly refreshed its product line.

Why this matters for product leaders:

Warby Parker proves that a single, resonant value proposition can survive. It even thrives through tariffs and inflation. The $95 anchor creates a clear mental model for shoppers. It forces every internal decision to prioritize affordability.

How Warby Parker Enables Growth

Vertical integration is a cost shock absorber. The company owns two optical labs in the U.S. It runs its own stores and e-commerce. This allows Warby to bypass some tariff impacts. It can move production out of China. It can also iterate on SKUs faster than rivals.

It uses a price ladder instead of across-the-board hikes. About 60% of frames still start at $95. But progressive lenses, blue-light filters, and new designs move customers into higher price points. This has lifted revenue per customer every year since the 2021 IPO.

The "forever price" creates a psychological moat. Holding the $95 price for 15 years builds trust. Even minor increases on premium SKUs feel reasonable. Competitors raised their entry prices multiple times.

The company is expanding its demographic. It targets aging consumers who need pricier progressive lenses. This offsets softness among younger, more price-sensitive buyers. It broadens the total addressable market without changing the brand ethos.

Takeaways for Product and Design

Invest in a flagship offering that never changes. Innovate around the edges. Each new frame style or lens upgrade drives upsell. The flagship price anchors customer perception. This balances consistency and novelty to stay relevant.

What PMs Can Steal Today

Protect your base plan or entry SKU. Make your finance team justify any changes to it.

Build flexibility into your production. Own or partner on production to protect pricing.

Use anchoring to make premiums feel like upgrades. They should not feel like price gouging.

Segment by life stage and functional need. Progressive lenses are Warby Parker's pro tier.

Warby Parker's playbook shows that growth is not about volume at all costs. It is about defending the core promise. Then widen the margins around it. Do this with customer-aligned innovation.

At some point, Warby Parker's base price offering will go up. Nevertheless, with that 15-year track record, this is a successful case study of anchoring value prop around your core offering.

Yesterday, we explored how AI transforms medical understanding, informing patients. Today, let's examine where AI actually delivers results in clinical practice. New research from the Journal of Clinical Medicine maps the gap between hype and reality. Resonates well with my personal experience.

"The central challenge is evident: as AI tools become more sophisticated, our capacity to integrate them ethically, equitably, and effectively into clinical practice must evolve in tandem. This editorial explores the remarkable progress in AI-driven medicine, identifies critical gaps that hinder its full potential, and highlights the collaborative efforts necessary to build a patient-centered future."

Where AI Works Today

AI succeeds in four key areas right now (confining to this research, there are several more that I am aware of):

Imaging leads the pack. The FDA has approved over 600 AI medical devices, with 75% focused on radiology. IDx-DR detects diabetic retinopathy with 87% sensitivity. Aidoc flags brain hemorrhages 96% faster than standard workflows. These aren't experiments. They're deployed in hundreds of hospitals.

Drug discovery shows measurable wins. Atomwise identified new Ebola treatments in days, not years. BenevolentAI found baricitinib as a COVID treatment in four days. Insilico Medicine took a novel drug target to clinical trials in 18 months. Traditional timelines? Three to six years.

Clinical decision support reduces errors. Epic's sepsis prediction model alerts doctors six hours earlier than standard protocols. Johns Hopkins reduced readmissions by 26% using machine learning on discharge planning. Mount Sinai's deep learning system predicts acute kidney injury 48 hours in advance.

Administrative tasks get faster. Nuance's Dragon Medical reduces documentation time by 45%. Carbon Health's AI scheduling fills 30% more appointment slots. Prior authorization that took days now takes hours.

The Implementation Gap

But most AI projects fail. Here's why.

Data quality kills promising tools. Hospital systems average seven different EMR formats. Lab results use different units. Missing data reaches 30% in some fields. You can't train models on chaos.

Integration breaks workflows. Adding new software to 20-year-old systems isn't simple. Doctors already click 4,000 times per shift. New tools that add steps get abandoned. Memorial Sloan Kettering spent $62 million on IBM Watson before pulling the plug.

Regulations slow everything down. FDA approval takes 12 months minimum. HIPAA compliance adds complexity. Europe's MDR requirements doubled documentation needs. Each country has different rules.

Trust remains fragile. Only 38% of physicians feel comfortable with AI recommendations. Patients worry about privacy. Blackbox algorithms face liability questions nobody can answer yet.

What Actually Moves the Needle

Successful deployments share patterns.

Start with narrow problems. Olive AI began with insurance verification only. They process 5 million claims monthly. Broader platforms struggle. Focused tools ship.

Augment, don't automate. Pathology AI highlights areas of concern, but pathologists make diagnoses. Radiology AI prioritizes urgent cases, but radiologists read them. Tools that support beat tools that replace.

Measure patient outcomes. Accuracy means nothing if patients don't improve. Ochsner Health's hypertension AI showed 71% blood pressure control rates versus 31% standard care. That's what matters.

Build learning systems. Cleveland Clinic's models improve monthly from feedback. Static algorithms become obsolete fast.

The Next 12 Months

Multimodal models will combine lab results, imaging, and notes simultaneously.

Personalized predictions will replace population averages. Your genetic profile, lifestyle, and history will generate custom risk scores.

Real-world evidence will separate winners from losers. Deployment data from thousands of hospitals will reveal what actually works.

The winners won't be the most sophisticated algorithms. They'll be the ones that fit into Tuesday morning rounds without adding friction.

Ben Thompson's latest piece hits on something crucial: when computing paradigms shift, yesterday's winners often become tomorrow's strugglers.

The risk both companies are taking is the implicit assumption that AI is not a paradigm shift like mobile was. In Apple’s case, they assume that users want an iPhone first, and will ultimately be satisfied with good-enough local AI; in AWS’s case, they assume that AI is just another primitive like compute or storage that enterprises will tack onto their AWS bill.

The article is worth reading in full. The author is much more bullish on Google adapting well. Pushing the key points a bit further:

First, the winner's curse isn't just about technology - it's about organizational metabolism. Big winners, sometimes, don't just have the wrong tech stack; they have the wrong clock speed. Google's chaotic, college-campus culture that Thompson praises isn't just quirky - it's actually a survival mechanism. When you're moving fast and breaking things (to borrow from the old Facebook motto), you can pivot faster than companies optimizing quarterly earnings. Apple's methodical perfectionism and Amazon's operational excellence are strengths until the game changes faster than their planning cycles.

Second, we're seeing a new kind of paradigm shift - from deterministic to probabilistic computing. Previous shifts Thompson mentions moved us from batch to continuous, from desk to pocket. But those all dealt with deterministic systems where inputs produced predictable outputs. AI represents something different: systems that are fundamentally uncertain, continuously improving yet hallucinate, that require verification rather than trust. This isn't just a new platform; it's a new computational philosophy that makes previous optimization strategies obsolete.

Third, the real disruption might be in business models, not just technology. AWS's bet is that a necessary component of generative AI being productized is that models fade in importance, but what if the opposite happens? What if AI makes compute itself commoditized while model differentiation becomes everything? Apple sells premium hardware in a world where software was commoditized. AWS sells commodity compute, where software is differentiated. Both models break if AI inverts these assumptions.

Fourth, there's a geographic dimension. Silicon Valley companies dominated the mobile and cloud eras partly because they clustered together, sharing talent and ideas. But AI development is more distributed, with strong players in London (DeepMind), Paris (Mistral), and elsewhere. The winner's curse might hit not just companies but entire ecosystems that assume proximity still matters the way it used to.

Finally, the timeline for disruption is compressing. It took Microsoft over a decade to miss mobile. It took Meta about five years to nearly miss AI (saved by pivoting hard into open source). The next paradigm shift might give incumbents months, not years, to adapt. This acceleration means the winner's curse isn't a slow disease anymore - it's a sudden cardiac event.

Thompson could be right that Google successfully navigated one paradigm shift, and is doing much better than I originally expected with this one. But that might be precisely because Google never fully won the mobile era - they stayed hungry while Apple counted iPhone profits. Sometimes the best position for the next race is second place in the current one.

The real question isn't whether Apple and Amazon will adapt to AI - it's whether any company can maintain dominance across multiple paradigm shifts when those shifts are accelerating. The winner's curse might just be getting started.

TDD

I've been playing around with various vibe-coding tools. While working on building this blog using Astro, I asked Claude Code to use a test-driven development (TDD) approach.

Bingo! It just built a whole test bed and followed the TDD approach for every new feature that's built.

I liked Bolt.dev, but since I started using Claude Code, it's a completely different experience. Bolt can still provide compelling prototypes. That, along with Claude Code, takes it to the next level with its much more intelligent implementation.

I've also asked it to make it a practice without me reminding it of TDD every time. For that, Claude inserted the following into its config file:

Eventually, I ended up with 90 tests and Claude executing them as I work on improving the functionality.

Be careful with overfitting

Apart from this, I had issues with rendering images on the homepage. It took me a couple of iterations to get it right with Claude. Sometimes, you see the model overfitting to your precise rules. Even to an extent, it hardcodes certain logic, in this case, the image file name. Being with it, nudging, and providing additional context usually fixes the issues.

So far, so good! I can clearly see where we are heading ... a world of exponential efficiencies from where we are.