How AI is shaping the future of software from creation to consumption 

Often in business, your future depends on being right about one thing. And today, that thing is generative AI. The largest winners in the next decade will be those who steer decisively into this shift. The modest winners will hedge their bets. And the losers? They’ll misread the stars and land far from their intended destination.

The maker landscape has transformed. With tools like Claude Code, Cursor, Windsurf, and Vercel v0, launching small apps now costs little more than a few API calls. This mirrors how word processors and digital cameras revolutionized books and photography.

With this decline in cost, software creation has exploded, facilitated by new platform capabilities from foundational model providers like Anthropic and OpenAI, IDE vendors like Replit, and hosting platforms like Vercel. These aren’t "just for developers"—they are idea-to-software pipelines, reaching beyond the traditional boundaries of engineering.

Whereas the cost of producing software previously limited these pipelines to de-risked ideas, this new creative flood is a brute force search for anything interesting. As Alan Kay observed, “a large enough change in quantity introduces a qualitative change” in behavior. Replit’s latest newsletter headline is “Create 3D Games in minutes.” Yes, most of this early wave will be throwaway, such as the iPhone-as-beer-glass gimmicks, but the breakthroughs are already starting to happen.

LLMs have helped to spawn 3,000 new entries in the marketing technology space in the last 18 months, and eventually, more new software than we can comprehend.

This tidal pool of software is also a breeding ground for competition over defensible niches adjacent to Customer.io. Hungry for revenue and market footing, these tiny products are eating into high-margin pain points in sales and marketing. Clay, which started as a tool for sales development reps to enrich outbound leads in a table, is now a market-leading solution for agentic workflows that gather data for a broad swath of use cases, acting as a top-of-funnel customer data platform (CDP).

Leveraging AI as makers, downmarket alternatives like Loops will have the opportunity to pursue their own path of expanding use-cases with a very small team. Even newer entrants will take advantage of coding strategies that exceed what we have today. The next captivating point solution will be built by a single engineer in less time and require less capital. What passed for advanced code generation two years ago (“code completions”) is being surpassed by chat-based programming, which, with enough infrastructure, will give millions of knowledge workers the ability to make atomic apps, as Steve Yegge charts.

When the market’s ability to deploy LLMs to generate code also begins to scale, competition will arrive from directions we wouldn’t expect, faster than we predicted. Case in point, by integrating with Webflow, Clay has followed demand from SDR lead generation support to scale website content personalization.

Some of this software will start as startups. Others will start from a marketer or salesperson scratching at a problem they have personally, without any initial commercial intent. As Scott Brinker shares in “Changes in the MarTech Landscape":

This army of inventive apps will eventually mature into a swarm of solutions in search of more use-cases, distribution, monetization, allies, and incumbent weaknesses.

If it’s true that a new engine of zero-to-one software production has been ignited, we should expect the bottleneck to shift to our downstream ability to publish, host, filter, surface, share, and discover these valuable artifacts. In other words, AI will drive demand for new platforms that handle software, in the same way the carefully hyperlinked, hand-crafted blogosphere yielded to the arrival of social media giants’ commoditized units of content and feed algorithms.

Regarding software-as-media, today’s publishing platforms are very high-friction, resembling the creative contexts for writers and photographers pre-web and pre-iPhone. Sit down, load up your IDE, give your app a name, make sure it belongs to a company with a business address, add a logo, upload screenshots, submit for review, wait, respond to the gatekeeping authorities, and hope your creation gets published.

Once you’ve tasted software creation via prompt ("vibe coding"), this multi-day journey of permission-seeking is obviously obsolete and excruciating. Vercel’s v0 platform has a single button for publishing and sharing your creation with the world, with options borrowed directly from YouTube’s interface.

Just as “who is and isn’t a photographer” became a moot question after iPhones and social media, the distinction between who is and isn’t technical will also fade; software’s Instagram era is here. A few will catch fire and go viral for every thousand, ten thousand, or one million throwaway creations. Like their social media counterparts, their primary asset will be attention, not intellectual property. These flashes will be used to funnel traffic into fledgling startups of one or two people. Rather than YCombinator-or-bust, new off-ramps for monetization will emerge in the form of long-tail software influencers and streamer equivalents. Power law dynamics will separate the Mr. Beasts from the market floor, but the “someone makes how much for playing World of Warcraft?!” shock will repeat itself too, but for software that exists as art, entertainment, one-time use, or potentially commercial.

The competition among makers will be intense. And while these idea-to-side-hustle stories will fuel new interest, the largest commercial winners will be the Instagrams, Twitches, and YouTubes of software. New portals like Websim.ai will sift the slop for the gems, and someone will figure out how to profit from doing so.

As makers gain power, end users of new software tools and platforms are gaining new abilities—and raising their expectations. Once the froth of experimentation fades, better UI patterns will emerge, making old interactions feel outdated. Patience will wear thin for interfaces that could “obviously” be improved or bypassed with LLMs. Products in commoditized markets—where switching is easy—will feel this pressure first. Over time, ChatGPT and its alternatives will make natural language features a baseline expectation. Prompts and agents will give users powers once reserved for those who mastered keyboard shortcuts and macros.

With AI-infused software on the rise, we’ll need not only fun features for discovery and automation, but also critical ones that ensure safety, privacy, fraud prevention, and fairness. Deepfakes will move beyond images and text into video, apps, and code. Some will come from agentic bad actors far outside regulated platforms like Capterra or TrustPilot. This creates opportunities for platforms to add tools to run code, detect fraud, and recommend trustworthy options.

Just as cloud SaaS drove demand for integrations, users will push for ways to connect small but essential apps into cohesive workflows. Tools like Zapier, Make, and n8n will face mounting pressure as supporting endpoints for thousands of SaaS apps becomes unsustainable. New layers like MCP are the early edge of formalizing a viable approach. hint at a more viable, formalized approach.

Because LLMs accelerate the average user's ability to express (and get) what they want from a product, we should also expect that end user to form their own opinion more rapidly on the best way to create and extract that value.

We can break this into phases by turning to our friends Dunning and Kruger.

• During Phase A, users are unskilled and lack confidence. This makes people hungry for an expert’s opinion to fill the gap: “show me the way.”
• During Phase B, users briefly possess a strong opinion borrowed from their minimal experience or environment (ex., the latest book, blog post, or conference speaker at the industry conference). They are ripe to buy a product that matches this newfound opinion.
• During Phase C, users descend into an “I don’t know” valley where they form their own opinion based on their earned knowledge while being skeptical of others. Considering opinions as fads, trends, or shoes (wear what fits the occasion), these buyers are the most satisfied by products offering depth and flexibility.

By accelerating the acquisition of skills and knowledge that shape opinions, LLMs will shorten Phase A, atomize Phase B, and reward those best equipped to support Phase C.

Just as today’s voter is shaped by thousands of personalized content impressions, not a few Cronkites, Brokaws, or Jon Stewarts, tomorrow’s buyer will be shaped by abundant software, not a handful of articles, touchpoints, or recommendations. They might take goals from a team meeting and use a custom GPT to generate campaign ideas, a mobile-specific calculator to reverse engineer target metrics, and a competitor’s simulator to model costs and diminishing returns. They might use all this to describe and design the growth engine of their dreams. Maybe a Granola.ai agent will do it from a voice recording. Maybe someone on their team will use a Replit that deploys agentic developers to code and host an MVP matching their exact specs. Maybe they’ll love it. Maybe it won’t work, and they’ll turn to Customer.io.

Buyers now resemble creators, using GPTs to ideate, apps to simulate, and agents to build. Whether they start with Customer.io or find us later, they’ll arrive with formed opinions. The question isn’t “are they technical?” but “are we flexible enough to serve them?”

Beauty, speed, and performance still matter—and always will. But the new gold standard is adaptability. No matter how svelte the UI or smart the settings, “You can favorite your top 10 stations” will lose to a Spotify playlist piped over Bluetooth the moment the car is entered.

We’re building an interface that will be judged by whether it has CarPlay. Delightful design now means recognizing and supporting the customer’s existing preferences, not baking in our own. The best products won’t feel like platforms, but like accelerators of what the customer already wants to do.