How the AI Wave Differs from Past Tech Revolutions
Artificial intelligence is often hyped as “the next big thing” – but is it truly a paradigm shift, or just better software? In a recent Acquired podcast interview, industry veterans Bret Taylor and Clay Bavor (co-founders of the AI startup Sierra) argue that today’s AI wave is fundamentally different from previous technology cycles. Taylor and Bavor draw on their careers at Google, Facebook, Salesforce, and beyond to compare AI’s trajectory with earlier revolutions like PCs, the internet, and mobile. They also describe how AI is spawning new business models (such as Sierra’s outcome-based pricing) and forcing companies to rethink how software is built, sold, and integrated. Below we explore the key insights from their conversation – from blistering adoption curves to autonomous “agent” software and its societal impact – with added context (including examples from Team Ignite’s portfolio) on how this AI wave both parallels and departs from prior tech transformations.
Accelerating Adoption: From Decades to Days
One striking difference is the unprecedented speed at which AI is being adopted compared to earlier technologies. Past innovations often took years or decades to penetrate markets, especially when physical infrastructure was involved. For example, it took decades for the landline telephone to reach 50% of U.S. households, a milestone that cell phones achieved in just five years by the 1990s. The World Wide Web first emerged in 1991, yet only 10% of the world’s population was online by 2002 – roughly a decade of gradual growth to hit that mark. By contrast, recent digital platforms can explode in popularity almost overnight. Smartphones accelerated this trend: in the span of about a decade, they went from novel gadgets to near ubiquity, with roughly 5 billion people (60% of the world) owning a smartphone by 2024. With billions of users already connected via the internet and mobile devices, new services today have a ready-made distribution channel unimaginable in earlier eras.
AI’s breakout moment vividly demonstrates this compression of adoption time. OpenAI’s ChatGPT, launched in late 2022, rocketed to 100 million users in just two months, making it the fastest-growing consumer application in history at that time. (For comparison, it took TikTok about 9 months and Instagram 2.5 years to hit 100 million users.) The viral chatbot reached 1 million users in only 5 days. This record was then blown away in July 2023 by Meta’s Threads app, which – piggybacking on Instagram’s network – amassed 100 million signups in 5 days. While Threads’ engagement dropped off after the initial hype, the key point remains: the plumbing of the modern internet (global connectivity, app stores, social media, etc.) enables new products to achieve scale at blistering speeds that simply weren’t possible in earlier tech waves.
This layered build-up of infrastructure is something Taylor highlights: each wave laid groundwork for the next. The PC era put computers on hundreds of millions of desks. The internet connected those computers, then mobile put a networked computer in everyone’s pocket. Thanks to these foundations, an AI service can reach a huge audience almost instantly via existing devices and networks. “When you make something like ChatGPT, you can go from zero to 100 million users faster than any technology in history because of all the technology that came before,” Taylor notes. In other words, AI’s adoption curve isn’t an isolated phenomenon – it’s riding “on the coattails of all of that infrastructure buildout” from prior waves (internet, smartphones) that drastically lower the friction for new software to spread.
Looking ahead, one might ask if there is an upper limit to this acceleration. Can anything top Threads’ 100 million in five days – say, a billion users in 24 hours? The Acquired hosts joked about that scenario. It remains unlikely in the near term, simply because even viral products need some awareness and desire from humans to sign up. Yet Taylor and Bavor wouldn’t rule it out in the future. If AI itself drives discovery and onboarding (imagine personal digital assistants automatically finding and using new services on our behalf), the concept of “distribution” could transform. We may see applications or agents that rapidly propagate because machines, not people, handle the bulk of onboarding. In short, the adoption bottleneck keeps shrinking – from hardware deployments (slow), to human awareness and clicks (faster), to potentially AI-driven automation (instantaneous). The AI wave is compressing timelines to levels that force us to rethink how quickly society and businesses can absorb new technology.
AI Agents: Software That Acts, Not Just Helps
Beyond raw user numbers, Taylor and Bavor emphasize a more qualitative break from past technology: AI isn’t just a tool to boost human productivity – it can autonomously complete tasks. They observe that earlier software generally served as a tool operated by people. Whether it was desktop software or mobile apps, the technology has been an instrument that might make you more efficient or effective, but you are still ultimately doing the work. AI changes that equation. Modern AI systems (especially those powered by large language models) can reason, decide, and take actions in ways software never could before. This gives rise to the notion of AI agents – programs that have a degree of agency to carry out goals for you, rather than just assisting you in the process.
Both co-founders are so bullish on this concept that they named their company Sierra around it: “Every company needs an agent,” their launch manifesto declared. By “agent,” they specifically mean an AI system that can interact naturally (e.g. via conversation) and execute end-to-end tasks in place of a human employee. In Sierra’s case, the focus is on customer experience agents – essentially AI-powered customer service representatives and sales/support personnel. Instead of a static FAQ chatbot or a tool that helps a call center rep, Sierra’s agent is the rep. It can converse with customers, troubleshoot issues, and actually resolve requests by integrating with backend systems.
Crucially, these agents go beyond answering questions – they can perform operations. For example, Sierra built an agent for SiriusXM that can handle the entire process of transferring a car radio subscription: the AI not only walks the user through it, but actually sends a signal to a satellite to activate the service in the new vehicle. Another Sierra agent for a home security company can diagnose an alarm panel’s low-battery alert and automatically ship the customer a replacement battery. One e-commerce client even had their AI agent process a warranty claim for sandals – the bot parsed the customer’s message, verified a photo of the broken product, and initiated a shipment of new flip-flops without any human involvement. These examples thrilled the Sierra team (who literally cheered when the first autonomous warranty was processed) because they proved the technology could carry out real-world business processes, not just chat.
Sierra is not alone in pursuing this new paradigm. Many startups are now building specialized AI agents to autonomously handle work in specific domains. For instance, Team Ignite’s portfolio includes Feather, which provides voice-based AI agents for the lending industry. Feather’s platform automates borrower interactions like lead follow-ups and payment reminders using voice and chat bots that work 24/7, essentially replacing the need for human call center staff for routine communications. By continuously calling customers and handling inbound queries around the clock, Feather’s AI agents function as always-on loan officers. Likewise, another Ignite company, Quandri, deploys “digital workers” for insurance agencies – AI bots that take over the time-consuming, repetitive tasks in policy administration and paperwork. These digital assistants can, for example, process renewal forms or update client records, freeing human employees from drudgery. In each case, the software isn’t just a tool used by a person; it’s acting in lieu of a person to complete the job itself.
In AI circles, a lot of terminology is floating around for these concepts (“AI agents,” “agentic AI,” etc.), and the buzzwords evolve rapidly. Taylor quips that tech lexicon often goes through faddish phases – “remember when we said ‘information superhighway’?” – and Bavor notes how quickly “prompt engineer” came and went as a job title. But they do believe “agent” will stick as the long-term term of art, just as we settled on “app” in the mobile era. An agent will simply be what we call a piece of software that autonomously handles a certain job or interaction (much like an “app” is just an application on your phone). The novelty will wear off, and agent will become a bland descriptor rather than a flashy marketing term – a sign that the technology has matured into an assumed part of life.
For now, however, AI agents are a new paradigm, and businesses are just learning how to deploy them effectively. One immediate implication is a rethinking of software success metrics. Traditionally, if you bought a software tool, it was up to your team to use it well and achieve outcomes. With agents that directly produce outcomes, the measure of success becomes what they accomplish rather than how many people use them or how many tickets they process. This leads to a transformative idea in enterprise software: outcome-based pricing.
Outcome-Based Pricing: Software that Gets Paid for Results
If an AI agent behaves like an autonomous worker, why not pay it like one? That’s essentially what Sierra’s business model does. Taylor and Bavor chose a radical pricing model for their startup: Sierra charges customers only when its AI agents successfully resolve an issue or complete a task. In other words, the software’s fee is contingent on delivering the outcome the customer cares about – for instance, solving a support ticket, closing a sale, or fulfilling an order. If the agent tries but fails (say it has to hand off to a human or the problem goes unresolved), the company doesn’t pay for that attempt.
This is a sharp break from the traditional enterprise software models of the past ~40 years. In the old on-premise days, companies paid upfront licenses (and annual maintenance) for software regardless of how much it was used or the results. The 2000s brought Software-as-a-Service (SaaS), which shifted to subscription fees (often per user or per seat). SaaS aligned cost more to usage, but still not to outcomes – you might pay for 1,000 seats of a CRM system, but whether your sales actually increase after implementing it is your problem, not the vendor’s. As Taylor puts it, “If you install a CRM and your sales don’t improve, no one blames the software – it’s on your team.” The vendor’s obligation in SaaS is to provide a working tool; achieving the business result is largely up to the customer.
AI changes this dynamic because the software can be directly responsible for the outcome. Sierra realized it could flip the model so that the vendor shoulders more risk and responsibility, charging only for what works. For example, Sierra’s customer ADT (the home security firm) pays per fully resolved support call that Sierra’s AI handles, and a major retailer pays Sierra a commission whenever the AI agent upsells a customer on a premium delivery service (just as they would pay a human sales associate a commission). In essence, the AI agent is treated like an employee or contractor – if it gets the job done, it gets paid. If not, no pay.
From the customer’s perspective, this model is highly attractive. It turns software spending into a variable cost tied to success, and it guarantees ROI by definition (you’re only paying when value is delivered). It’s also far more transparent than convoluted SaaS pricing tiers. As an analyst at Verdantix observes, outcome-based pricing aligns perfectly with the rise of autonomous AI agents: “As AI agents’ ability to autonomously complete tasks increases, user numbers no longer correlate to software value… There are already examples of B2B software firms with outcome-based models – Sierra, for example, only charges clients for customer tickets resolved by its AI agents, rather than a subscription or usage fee”.
However, this model also demands much more from the vendor. Sierra is essentially taking on the delivery risk that the customer’s employees or outsourced agencies used to bear. It forces a deeper partnership relationship. Sierra can’t just drop off a piece of software and say “good luck”; they are incentivized to actively ensure the agent performs over time (since their revenue depends on it). Taylor notes that this blurs the traditional divide between “pre-sale” and “post-sale.” In legacy software, once the sale was closed, the vendor’s job was mostly done aside from support and hoping the customer renews next year. In an outcome-based world, the real “sale” isn’t closed until the software proves itself through results. This drives Sierra to work closely with customers on implementation, tuning, and continuous improvement of the AI agent. The success of both parties is literally tied together.
Interestingly, this harks back to how we compensate human workers or even how ad platforms charge for performance. In online advertising, models evolved from paying per impression (CPM) to paying per click (CPC) to even paying per conversion or sale. Each step shifted more risk onto the seller but also aligned value more tightly with outcome. AI enables the same shift in software. It may start in domains like customer service or sales where outcomes are clearly measurable (e.g. issue resolved, transaction completed), but if successful, it could inspire broader changes in enterprise software pricing. We may eventually expect software vendors to put more “skin in the game” and charge for what their product actually achieves rather than flat fees for promised capabilities.
Notably, even before true outcome-based deals take off, many software companies are moving toward flexible, usage-linked pricing – a trend accelerated by the AI era. Cloud infrastructure and API providers, for example, bill based on consumption rather than seats, reflecting how value scales with actual usage. Startups are emerging to support this transition. Outcome-based pricing can be seen as the next evolutionary step on this spectrum – pushing the pay-for-value concept to its logical extreme. Taylor goes so far as to predict this could be the next major business model disruption in enterprise tech. Just as the SaaS shift in the early 2010s upended how software was sold (and made winners of those who adapted, while others struggled), the AI era might force another shake-up. It won’t happen overnight, but if customers start demanding outcome-based arrangements, software firms will need to adjust. At the very least, as AI handles more of the actual work, metrics like “seats” or “active users” make less sense as the basis for pricing. The value might be delivered by one AI agent doing the work of 100 employees – charging per seat in that scenario is absurd. Either you charge based on some notion of consumption (compute usage, API calls, etc.) or you charge by outcome delivered. Sierra’s bet is that outcome-based deals will win in the long run because they align incentives and simplify the value proposition: pay for results, not effort.
Broader Implications: Jobs, Identities, and an Agent-to-Agent World
The rise of AI agents and their breakneck adoption has far-reaching implications, which Taylor and Bavor discuss with a mix of optimism and caution. At the highest level, Taylor frames AI as the next step in a historical pattern of turning scarce resources into plentiful utilities. Previous industrial revolutions did this with energy (e.g. electricity made power abundant and cheap) and with information access (the internet democratized data). AI, he suggests, is doing it with intelligence or expertise. “We’re essentially taking expertise and making it a commodity,” he says – potentially giving everyone access to the knowledge and skills that only specialists or the wealthy could once obtain. Imagine universal personal tutors, mental health counselors, legal aides, and so on, available on-demand to anyone with a smartphone. In Taylor’s view, that’s a profoundly positive development: many forms of advice and services could become cheaper or free, lifting up people who historically had no access to them. “What a cool thing that we’ve made this universally accessible,” he remarks about AI’s ability to scale expert-level help to the masses.
This democratization of expertise is not theoretical – it’s already underway in various fields. Knowtex (another Ignite company) gives every physician their own AI assistant: Knowtex’s voice AI system listens to doctor-patient conversations and automatically generates detailed clinical notes and billing codes, saving doctors over two hours per day in paperwork knowtex.ai. By handling tedious documentation, it not only boosts productivity but also addresses physician burnout. In effect, each doctor gets a tireless scribe and billing specialist for free, courtesy of AI. These examples show how capabilities once limited to experts or large teams (specialist medical advice, or a personal admin staff) can be scaled out to many via intelligent software.
At the same time, when something scarce becomes abundant, society goes through a rocky transition. Taylor draws an analogy to how people in developed countries no longer worry about basic needs like food or lighting in the way they did a century ago – those became utilities, freeing us to focus on higher-level pursuits. AI could likewise elevate humanity’s “job description” to more creative and meaningful endeavors once rote cognitive labor is offloaded. But in the near term, there’s dislocation. Individuals feel existential anxiety when the very thing they’ve built their career and identity on (their intelligence, expertise, creative skill) suddenly seems replicable by a machine. Notably, it’s knowledge workers and tech professionals who are feeling this most acutely right now – “the people building the technology are disrupting their own profession,” Taylor observes wryly. This is unlike past automation waves that primarily affected manual or repetitive labor. A recent study by OpenAI and economists found that around 19% of U.S. workers could see at least 50% of their job tasks impacted by AI, and higher-wage occupations actually face greater exposure on average than lower-wage jobs. In other words, white-collar jobs are squarely in the crosshairs of this wave.
How this will play out in terms of employment is an open question. Bavor and Taylor are optimistic that, as with past revolutions, new jobs and paradigms will emerge. In their view, AI is a powerful “force multiplier” for talented people. They recount how their engineers at Sierra use AI coding assistants (“Cursor,” in this case) to write a large percentage of their code, effectively doing the work of many coders and allowing a small team to achieve outsized productivity. Rather than replace their engineers, it augments them – though it does change the nature of the work. Sierra even instills a cultural practice of “fix the machine, not just the output”: if the AI writes faulty code, the engineer’s job is to improve the AI’s context or instructions so it will do better next time, not merely fix the bug manually. This mindset treats the AI as an intern or collaborator that can continuously learn, rather than a one-off tool. It’s a glimpse of how work might evolve when human–AI teams are the norm.
On a broader scale, we’re already seeing AI assume many of the junior-level or routine tasks across industries, which in turn is redefining roles. In the hiring and HR arena, for example, AI systems are screening candidates, scheduling interviews, and even conducting initial Q&A chats with applicants. A platform like Humanly can “engage and interview 100% of your talent pipeline” via AI agents humanly.io, ensuring every applicant gets a response and basic evaluation without consuming a recruiter’s time. That means recruiters and hiring managers can focus on the higher-order work of building relationships and evaluating fit, rather than slogging through hundreds of résumés or scheduling emails. In customer support, AI agents are handling the majority of tier-1 queries; in sales, AI can qualify leads and even negotiate basics of deals. Rather than eliminating these jobs, companies are finding that the human workers in these roles are now managing and enhancing the AI-driven processes. The work shifts toward oversight, exception handling, and more complex problem-solving. As Taylor puts it, we’ll still need people to “own the outcome” – but the path to that outcome might involve supervising fleets of digital helpers.
One fascinating area to watch is how consumer-facing AI agents could transform markets and daily life. Bavor and Taylor discuss a future where your personal AI assistant talks to other AI agents on your behalf. For example, imagine your shopping agent automatically scouting various vendors’ agents for the best price and terms on a product – essentially negotiating deals for you. In fact, multi-agent experiments are already demonstrating this concept. Researchers at Cognizant built a prototype where a “consumer agent” interfaced with networks of business agents to plan a vacation, haggling over hotel rates and amenities with different travel providers’ AI representatives. The AI agents went back and forth, offering discounts and improved options, all without human intervention – a literal machine-to-machine marketplace. The consumer’s agent was instructed to operate in the user’s best interest, while the business agents served their companies, and they dueled it out just as humans would in a negotiation.
This leads to some mind-bending second-order effects. If agents are talking to agents, what happens to traditional marketing and sales funnels? Taylor posits that the internet’s current structure – where companies spend heavily on ads to influence human buyers (demand generation) and optimize their websites or search rank to be discovered (demand fulfillment) – could be upended. In a scenario where your personal AI agent simply asks another AI to find the best product for you, the human’s brand awareness or emotional advertising might not factor in much. Companies might need to appeal to algorithms (perhaps by offering favorable API interfaces or machine-readable specs) as much as to people. It could shift power toward platforms that host consumer agents (imagine if a future Alexa or Siri becomes the trusted broker of most transactions). Brands, on the other hand, will fight to avoid being commoditized in an agent-driven ecosystem – a topic the co-founders admit is hard to predict but likely significant. “Which companies will have their own agents and enough brand equity to attract customers directly, and which will be aggregated behind personal agents?” Taylor asks. Just as some retailers depended on Google or Amazon for discovery in the web era, businesses in the agent era might depend on placement in AI assistant recommendations.
Another implication is organizational. Large enterprises will need to adapt their culture and processes to leverage AI – which is often more challenging than the technology itself. Taylor, having led teams of thousands, notes how tough it can be to roll out new ways of working (as simple as getting everyone to use a new tool). Startups like Sierra have the advantage of a blank slate: they mandate an “AI-first” approach internally (e.g. all employees are expected to use AI tools like ChatGPT for research, coding, content drafting, etc., to amplify their output). Big companies will have to overcome inertia and re-skill workers to effectively collaborate with AI. Those that do can see huge efficiency gains; those that don’t risk falling behind more nimble competitors. The absorption rate of AI – how fast organizations and individuals can assimilate these capabilities – may end up the gating factor on how quickly the full potential of the technology translates into economic productivity. “The technology might be racing ahead, but society has its natural rate limiters,” Taylor observes, citing factors like regulation, adaptation time, and trust.
A New Paradigm, Not Just “Better Software”
AI’s current wave combines a rare set of attributes: historically rapid adoption, a step-change in what software can do (autonomous reasoning and action), and disruptive business economics that challenge old models. Taylor and Bavor argue that while it’s easy to get caught up in buzzwords, the big picture is that this is a genuine paradigm shift. “I think we’ll look back on this era as an inflection point on par with the advent of the internet,” says Taylor. That is a bold claim – recall that the dot-com boom was dismissed as hype when it crashed, yet two decades later internet companies utterly dominate the economy. Similarly, today’s AI exuberance includes plenty of “snake oil” and inflated expectations, but the signal amid the noise is real progress. The technology is already delivering tangible value (e.g. Sierra’s agents resolving over 80% of incoming customer issues on their own), and entire markets are adjusting to the shock.
Perhaps the most profound difference with AI is its intimate relation to human intelligence. Previous tech waves augmented our muscles (machines), memory (computers), or reach (networking), but this one touches our cognitive core. It forces us to ask, what is uniquely human when machines can think and act? Taylor is ultimately a technophile – he believes humans will harness AI to “increase our leverage” and focus on higher-level creativity, relationships, and endeavors that truly require the human touch. But he’s also humble about forecasting exactly how things will shake out in 5–10 years. No one fully anticipated how the internet would spawn new industries (and demolish others), or how mobile would change daily behavior. Likewise, AI’s second- and third-order effects – on employment, competition, social interaction, even our sense of self – are only beginning to unfold.
One thing is certain: AI is not “just better software.” It represents a new kind of technological force, one that demands new mental models. Business leaders must rethink how they price and deliver services when software can carry out actions autonomously. Workers must reimagine their roles in collaboration with increasingly capable machines. And all of us, as consumers and citizens, will need to navigate a world where intelligent agents become as commonplace as apps and websites – a world that will feel remarkably different from the last tech era. As the conversation with Taylor and Bavor illustrates, embracing this wave involves equal parts excitement and discomfort. But if history is any guide, those who adapt will find opportunity in the upheaval. After all, the ultimate promise of technology is to free us to do more of what humans do best. AI is simply testing how ready we are to take that next leap.
Sources:
Taylor, B., & Bavor, C. – Acquired (ACQ2) Podcast: “How is AI Different Than Other Technology Waves?” (Interview discussion, Aug. 2025).
Reuters – “ChatGPT sets record for fastest-growing user base – analyst note” (on ChatGPT reaching 100M users in 2 months) revero.com.
Social Media Today – “How Long Did it Take Apps to Reach 100 Million Users?” (Threads vs. ChatGPT adoption speeds) globalwellnesssummit.com.
Harvard Business Review – “The Pace of Technology Adoption is Speeding Up” (on telephone vs. cellphone adoption rates).
FedTech Magazine – “The Internet Has More Than 2 Billion Users…” (10% of world online by 2002).
BankMyCell / ExplodingTopics – Statistics on Global Smartphone Users (~4.88B in 2024).
Verdantix – “Rethinking Enterprise Software Pricing Models for the AI Era” (outcome-based pricing and Sierra example).
OpenAI Research – “GPTs are GPTs: An Early Look at the Labor Market Impact of Large Language Models” (Mar. 2023).
Cognizant AI Lab – “AI Agents Working for Consumers” (Jan. 2025 blog on multi-agent negotiation experiment).