AI Makes Opinionated Software More Important Than Ever

Research Rundown #143, plus new memos on Inworld AI, Clay, and more.

Jul 12, 2025

Research Rundown

The word “flexible” is often used as a selling point when talking about software. Zapier markets its 7K integrations, Notion touts 30K templates, and Twilio literally calls its contact center software “Flex” because it can do whatever customers want it to do.

But now we’re seeing an explosion of software creation as software enters what feels like a post-scarcity world. An ocean of “vibe coding” will produce dramatically more software than we’ve seen before. How can software companies differentiate when anyone with an internet connection and a niche idea for a feature can spin up a competitive product?

The answer might be abandoning flexibility in favor of offering software that is more “opinionated” and taking a stance on how exactly a product should work rather than letting users drown in a sea of flexibility. In one of our latest memos, we unpacked how Owner has done just that for independent restaurants. Owner’s CEO Adam Guild has argued that his product’s advantage comes from reducing options for customers and being “extremely opinionated” about its software, which he argues improves online ordering conversion rates and SEO ranking:

“All small business software before Owner was built to be maximally customizable. But Owner is optimizing for giving customers performance, not customizability. For example, restaurant owners can upload their logos and brand colors into the Owner website generator. But they can’t change the placement of CTA buttons, information hierarchy, or order conversion flow.”

From restaurants to project management, there are dozens of use cases that are going to be flooded with software. With AI, it’s never been easier to create software, but it’s also never been harder to find the right software. Opinionated software historically has been rarer and not always successful. But in an AI-fueled world, it might be the main way to differentiate the slop from the ideal software.

What Is Opinionated Software?

The idea of opinionated software is usually attributed to David Heinemeier Hansson (DHH), co-founder of 37Signals, which released the full-stack framework Ruby on Rails and project management software Basecamp. He coined the term in 2006 in a book he co-authored called “Getting Real.” In a chapter titled “Make Software Opinionated,” he argues that software makers should take sides when it comes to their products:

“Some people argue software should be agnostic. They say it’s arrogant for developers to limit features or ignore feature requests. They say software should always be as flexible as possible. We think that’s bullshit. ****The best software has a vision. The best software takes sides. When someone uses software, they’re not just looking for features, they’re looking for an approach. They’re looking for a vision. Decide what your vision is and run with it.”

Opinionated software isn’t just about how products are designed. It’s also about the workflows the product supports, and what the “right way” to do a job is. All the products DHH and 37Signals have made exemplify this philosophy.

Ruby on Rails, a full-stack, open-source web-application framework, prepackages the folder structure, database connections, and code templates to speed up developer onboarding and reduce the amount of time spent configuring every detail. Basecamp, a project management app similar to Asana, famously doesn’t have Gantt charts despite endless requests for them because it is “inconsistent” with what 37Signals thinks is the right approach to project management. With its email service HEY, all messages funnel into just three fixed streams, and users are forced to screen every new sender the first time they send a message.

Modern Use Cases for Opinionated Software

Since DHH’s book came out, there have been some clear examples of software companies adopting the same approach to building opinionated software:

Linear: Pushes teams to fixed cycles and its default status pipeline
Superhuman: Emphasizes inbox-zero, and optimizes for keyboard shortcuts
Arc (The Browser Company): Tried to reimagine browsing around “spaces” and a sidebar tab library, though they are now pivoting
You Need A Budget (YNAB): Forbids forecasting money you don’t have, enforces zero-based budgeting
Roam Research: Emphasizes bi-directional linking and doesn’t allow folders
Looker: Requires the use of LookML, its proprietary language, and ad-hoc SQL queries are highly discouraged
Medium: No customization, had “Direction over Choice” as a core design principle
Owner: Websites are all standardized for performance

Some of these companies, like Linear or Owner, have put their opinions front and center in defining their products. Others aren’t so obvious, like social media. Gone are the days of fully customizable HTML-based MySpace profiles. Except for a profile picture, a bio, and some links, every profile on X, TikTok, and Facebook looks the same. The introduction of algorithmic social feeds introduced an even stronger opinion. “We know what you want to see better than you do.” While consumers seem to have put up with these opinionated frameworks for social media, it hasn’t been as successful in software.

Opinionated Software’s Mixed Track Record

Opinionated software sounds great. People typically build software precisely because they have opinions, so why be “agnostic,” as DHH says, when you can take a stance? But the problem with opinions is that people have to agree with you if you want them to be successful. And that hasn't always ended up working out for the most opinionated software builders.

Consider DHH’s companies: Ruby on Rails is a success, with nearly 5% of developers using it according to a 2024 survey. But it has fallen far behind other web application frameworks like Node.js (41%) and React (40%), which are known for being far less opinionated and more flexible.

Basecamp, founded in 1999, bootstrapped to an estimated $280 million of revenue by 2024. But that still lags behind newer competitors like Monday.com, which was founded in 2012 and did $1 billion in revenue over the last 12 months, and Asana, which was founded in 2008 and did $738 million in revenue over the last 12 months. HEY launched more recently in 2020 and had “tens of thousands of customers” as of January 2024, but that pales in comparison to Gmail’s 1.8 billion users or Microsoft Outlook’s 400 million.

Opinionated software companies are by no means a collection of failed products. Basecamp and HEY’s success is meaningful. Linear was valued at $1.25 billion, and Superhuman and Looker were both acquired. But in a world where software companies are generating billions of dollars of revenue and shaping whole industries, opinionated software success looks more modest. If you invested exclusively in companies over the last 10 years based on how opinionated they are, you probably wouldn’t have done very well.

The Struggle To Be Opinionated

So what’s going on here? Why is building opinionated software so clearly not the path of least resistance? In some cases, these opinions are being put forward in what are already notoriously known as “tar pit” ideas, like productivity or personal finance. In other cases, companies are competing with behemoths like Google or Microsoft, which is never easy. But there are other fundamental obstacles to putting forward opinionated software.

First, being too opinionated limits your ability to handle different types of customers. The biggest software products in the world are highly flexible. Microsoft Excel is used by over one billion people, has hundreds of use cases, and caters to individuals, small businesses, and large enterprises alike. Excel obviously has some design choices that enforce constraints, but its design philosophy seems to be “sure we can add a button for that,” which allows for endless customization. Similarly, Salesforce’s AppExchange and Atlassian’s Marketplace now both list thousands of apps, letting their products flexibly integrate with thousands of other features.

So Owner’s opinionated software may work for small restaurants, but moving upmarket and working with the largest restaurant chains may force them to become less opinionated to succeed.

Second, opinionated software may have a TAM that is inherently limited to the people who happen to agree with your opinion. Sometimes, opinionated software is just wrong about what the vast majority of customers want. Roam Research pushed hard on the value of bi-directional links, but most people still take notes the old-fashioned way. The Browser Company’s Arc built a cult following, but most people still use a more traditional browser, and now the company is pivoting to a new browser.

Some companies end up abandoning their opinionated approach to accelerate their growth, like Instagram did. As designer Naz Hamid wrote, Instagram was highly opinionated when it started by only allowing photo uploads and not allowing videos or disappearing stories. Then Instagram’s rapid growth and intense competition from Snapchat and TikTok in its market made it reconsider its product, and like every other social media company, it now has “Stories” and “Reels”. Contrast that with VSCO, which has remained opinionated in its social network by emphasizing photos and not having like counts, comments, or algorithmic feeds. Unsurprisingly, VSCO also does not have the billions of users Instagram does.

Third, there’s a difference between taste and data-driven opinions. Many of the companies we’ve described have “taste” driven opinions. Roam Research, YNAB, and Linear didn’t design their products on A/B tests of what works, but instead based on how they think workflows should work!

The way Owner and TikTok are opinionated is different and driven by data. Owner is making decisions based on which website templates perform best on Google SEO, not based on some opinionated view on how websites should look. TikTok’s algorithm constantly adjusts based on experimentation, not just “taste” for what videos might be good.

Data-driven opinions seem to work quite well, in part because they are actually “flexible” to whatever the data is showing. It’s taste-driven opinions that seem to struggle, because they are often inflexible, even if the data gives reason to believe the opinion is wrong.

AI Pulls Opinions To The Surface

The introduction of endless software via AI means that any feature can be reproduced with a prompt. As Akash Bajwa, an investor at Earlybird, explains, this means AI will make the opinions and tastes of the software creator the primary way to differentiate against competitors.

Enterprise software incumbents have traditionally won by having the most complex but also the most feature-rich and flexible platforms. But AI is making it easier than ever to build whatever features you want with a few simple prompts. If there are 20 different software tools all with similar features, then differentiation will have to come from how opinionated each tool is on the user experience, designs, and workflows. In other words, if you can do something with any of these tools, then which approach is the way you should do it?

One 2024 report from Battery Ventures argues that, as technology moats diminish and off-the-shelf infrastructure makes it easy to spin up any product, product design in the application layer of any product will become the “offensive force and key differentiator”.

Anu Atluru’s viral essay “Taste is Eating Silicon Valley” makes a similar argument:

“In a world of scarcity, we treasure tools. In a world of abundance, we treasure taste. Everyone’s software is good enough. Software used to be the weapon, now it’s just a tool. The barriers to entry are low, competition is fierce, and so much of the focus has shifted — from tech to distribution, and now, to something else too: taste.”

Where some people talk about taste, others talk about being opinionated. But they each point to a core idea: Knowing what to build is becoming just as important as how to build it. The craft or perspective with which the software is built will set it apart.

Ideas Over Execution

There’s a long-running debate about whether ideas or execution matters more in building startups. You often hear “ideas are cheap and execution is expensive,” but that argument is becoming less persuasive now.

AI is making execution easier and raising the importance of ideas, but it's also creating a lot of uncertainty about what workflows will look like in the future. Will we all be wearing AI listening devices that integrate with a personal OS? Will the browser look fundamentally different? Will the AI chat interface still be dominant in five years?

There’s no way to A/B test these, and being opinionated and right about the way software will work in the future will create a new set of generational companies.

Inworld AI is developing an AI-powered Character Engine to enhance non-player computer (NPC) interactions, distinguishing itself from traditional game development tools focused on rendering and physics. To learn more, read our full memo here and check out some open roles below:

Forward Deployed Engineer (AI Gameplay Engineer) - Mountain View, CA
Senior Frontend/NodeJS Engineer - Mountain View, CA

Clay is a Go-To-Market (GTM) software platform that provides data enrichment, workflow automation, and AI agents to allow companies to find, contact, and close customers more effectively. To learn more, read our full memo here and check out some open roles below:

Analytics Engineer - New York (Hybrid)
GTM Engineer (Sales Led) - New York (Hybrid)

GlossGenius operates in the beauty and wellness software market, offering business management tools tailored to independent beauty professionals and small salons. To learn more, read our full memo here and check out some open roles below:

Senior Software Engineer (Backend) - Remote (SF Bay Area) or Hybrid (NYC)
Senior Software Engineer (Frontend) - Remote (SF Bay Area) or Hybrid (NYC)

The Farmer’s Dog is an ecommerce company that sells fresh pet food subscriptions to dog owners in the US with a specific focus on premium products. To learn more, read our full memo here and check out some open roles below:

Software Engineer 3 (Growth and Acquisition) - New York, NY
Senior Product Manager (Acquisition) - New York, NY

Check out some standout roles from this week.

Canva | Sydney, NSW (Remote) - Senior Backend Software Engineer (Edge - Cloudflare), Senior Software Engineer (Platform Workflows), Senior Engineering Manager (FE - Editing Core), Senior Machine Learning Engineer (Content Review)
Anduril | Costa Mesa or Irvine, CA - Sr. Reliability Engineer, Air Defense (Software Product Manager), Sustainment Engineer, Senior Flight Performance Engineer, Circuit Design Engineer, Flight Software Engineer (Embedded C/C++)
Scale | San Francisco, CA - AI Infrastructure Engineer (ML Data Platform), Developer Productivity Engineer, AI Product Manager (Generative AI)

OpenAI’s Windsurf deal collapses as key talent heads to Google. OpenAI’s planned $3 billion acquisition of AI coding startup Windsurf has been called off, with CEO Varun Mohan, co-founder Douglas Chen, and key R&D staff instead joining Google DeepMind to work on agentic coding and Gemini. Google will license some Windsurf tech but won’t acquire the company. Source.
The browser wars are heating up, as Perplexity debuts Comet and OpenAI is soon to follow with its own browser. Perplexity rolled out its AI-native Comet desktop browser to its $200-per-month Max subscribers, bundling an agent that summarizes email, manages tabs, and navigates pages. It has been reported that OpenAI will unveil its own AI browser in the next few weeks, built on top of Chromium, Google’s open-source browser.
xAI debuts Grok 4 and $300 “SuperGrok”. Elon Musk’s xAI launched Grok 4 and a $300 per month SuperGrok Heavy plan. Grok 4 topped competitor models on key coding, science, and math benchmarks, and the ultra-premium tier gives early access to its highest performing “Grok 4 Heavy” model. In the days prior to the launch, though, xAI had to take Grok’s chatbot down after it started making anti-Semitic and other inappropriate responses.
Japan reboots nuclear fleet 14 years after Fukushima. Tokyo will restart 14 idled reactors and green-light next-gen designs to lift nuclear’s share of the power mix from 8.5% to 20% by 2040, amid soaring LNG prices and data-center demand. The policy shift shows how small modular reactors are gaining traction as a low-carbon energy source, yet public resistance and cautious regulators could still cloud Japan’s nuclear revival.
Google says AI now writes half its code. Google’s research blog revealed that AI suggestions now generate 50% of code characters, up from 25% two years ago. The company plans to layer in larger Gemini models and expand automation to testing and maintenance, highlighting how enterprise software engineering could move far beyond code completion.
ChatGPT tests “Study Together” tutor mode. OpenAI has enabled a “Study Together” tool for some ChatGPT Plus users, turning the bot into a Socratic tutor that questions students to guide discovery rather than handing over answers. This puts OpenAI in competition with education startups like Khan Academy, which we covered in our deep dive on how AI is impacting student learning.
AWS to debut AI agent marketplace with Anthropic. Amazon Web Services will open the storefront at its July 15 Summit in New York, partnering with Anthropic to let startups sell agents directly to AWS customers. The move positions Amazon against Google’s and Microsoft’s new agent stores and gives Anthropic a broader enterprise channel.
a16z quits Delaware, urges founders to follow. Andreessen Horowitz is moving its business from Delaware to Nevada, warning that recent Chancery Court rulings have eroded the state’s once-predictable business-judgment rule. By touting Nevada’s codified liability protections and reformed business courts, Andreessen Horowitz hopes to normalize a broader startup exodus from Delaware.
Kalshi releases another wild AI TV ad. Director PJ Accetturo says the 60-second spot was scripted with Gemini, rendered in Google’s Veo 3, edited in CapCut, and took only one week to make. Kalshi also released an AI ad for the NBA Finals last month, which took two days to create and cost $2K to make.
Jack Dorsey unveils ‘Bitchat’, a Bluetooth WhatsApp rival. The peer-to-peer app passes encrypted messages across a 300-meter mesh without internet or phone numbers. It launched on TestFlight, reaching its limit of 10K beta users within hours.
Meta buys 3% of Ray-Ban parent EssilorLuxottica. The €3 billion investment could grow to a 5% holding and comes after last month’s announcement that Meta will create AI-powered smart glasses with Oakley, another EssilorLuxottica brand. The "Oakley Meta HSTN" will feature a hands-free high-resolution camera, open-ear speakers, water resistance, and Meta AI capabilities.
OpenAI & Microsoft back an AI teacher academy. The National Academy for AI Instruction will train 400K US K-12 teachers over five years, funded with $23 million plus credits and tools to integrate generative AI into lesson prep and assessment. Microsoft will provide $12.5 million, OpenAI will contribute $8 million in funding and $2 million in technical resources, and Anthropic will add $500K.
Is the market sufficiently pricing AI progress? Dwarkesh Patel says no. MetaCritic Capital called Dwarkesh’s piece on the continual learning bottleneck a sign that the AI trade may have peaked. However, Dwarkesh chimed in and argued that even if AI 2027 is too optimistic, the market is still far from pricing in the transformative impact he expects over the next decade or two.

At Contrary Research, we’ve built the best starting place to understand private tech companies. We can't do it alone, nor would we want to. We focus on bringing together a variety of different perspectives.

That's why applications are open for our Research Fellowship. In the past, we've worked with software engineers, product managers, investors, and more. If you're interested in researching and writing about tech companies, apply here!