The last year and a half have been transformative (no pun intended) for the tech industry. The rise of Gen AI as a technology has been meteoric. It took about 7 years from the landmark Google research paper ‘Attention is All You Need’ in 2017 to ChatGPT becoming an overnight success at the end of 2022. The last 1.5 years have seen evolutions of technology, frenzy, hype, and real economic impact at an unprecedented pace. We have gone on from novel text generation/chat usage to multimodal and agentic use cases across the business landscape. As per a talk by Sequoia US, Gen AI companies have reached upwards of $3 Bn revenue run rate in just three years compared to a decade taken by Cloud software companies.
Even with this progress, the general consensus is that we are in the early stages of this technology shift and there is still a long way before more possibilities emerge and create a huge impact on the global economy. We believe that like any other technological shift in the past, AI is going to be deflationary and will generate a net positive economic impact.
We will discuss some key trends that keep us excited and form most of our thesis-building and brainstorming sessions.
“GPT 4 is the dumbest model that any of you will have to use by a lot.’’
This statement by Sam Altman really hammers the point home that we are still very early in this technology evolution. The pace of technological progress has been tremendous in the last year and a half. While in the start we had OpenAI as the major player in the proprietary model API space, now we have a bunch of companies competing closely with them. Meta joined the party a little late but totally upended the game by open-sourcing a very capable Llama series of models, now open source is a serious option for many enterprises and developers. In less than eighteen months, we have gone from GPT 3.5 scoring a 70 in MMLU benchmarking to Google Gemini now pushing towards human performance at 90 in the same benchmarks. The throughputs and context windows which are important measures for real-life applications have similarly improved tremendously, for example, OpenAI models have gone from a 4k context window (GPT 3.5) to 128k (GPT 4 Turbo) during this period, and we now have 1M context window models also come up. Along with the capabilities, models have improved a lot on cost and latency as well.
While there have been continuing improvements in LLM capabilities, we are now also seeing some signs of progress on totally different vectors as well. There has been progress across modalities (GPT 4-O is a leap in the multimodal direction), architectures (like Mixture of Experts) and hardware (innovation beyond GPUs). In such an environment of rapid technological progress, entrepreneurs/developers must take a long-term view while building applications, or else the risk of irrelevancy looms large with each tech jump.
“The king is dead, Long live the king”
‘Gaurav, what do you think, is SaaS going to be dead in the next 5 years?’ This was the question a VC friend who runs a large growth fund focused on enterprise technology asked me and my colleague Veenu on the sidelines of SaaSboomi. The irony of two SaaS-focused investors discussing this during the biggest SaaS event in the country was not lost on us. Provoking headline aside, we all agreed that this is the biggest transformation that is happening in the software world after the cloud. Software as built, consumed, and distributed will most likely change in the coming decade.
The biggest change that we believe, is that software will progressively move to outcomes. If you look at the history of software, it went from On-Prem one-time licensing to cloud (SAAS) as the customers started demanding a more value-based pricing model. The way the current software is architected and consumed is to be the ‘productivity enhancer’ for human workers where the pricing evolved to be per user (or per seat). In the last few years, customers have started pushing for more value out of the software which resulted in a move towards usage-based pricing (h/t-Togai) and SaaS value management (h/t- Spendflo). We believe that this trend will continue and software will move towards an ‘outcome-based’ approach. The cloud era was all about the software storing and moving information (data) on demand for the ‘intelligent’ human users; AI will also add another primitive of providing (some form of) intelligence on demand to the user.
2024 has seen a lot of discussions around AI Agents. While still nascent, we are now seeing agents really driving value across different use cases. Our portfolio company SuperAGI has been ahead of the curve and is at the forefront of this shift. Improvement in AI agents will make the capabilities of the ‘intelligent, automated software’ start encroaching on the services territory. Maybe SaaS does evolve into a productized services model, a real ‘software and a service’
“Today’s business models won’t last more than a decade”- Deepinder Goyal
We believe that AI is fundamentally changing the way technology is built, consumed, and distributed, which means the business models will evolve. Everyone is already seeing how the cost of building technology is becoming cheaper with tools like GitHub copilot. We are also now seeing initial signs of AI changing the way products/services are distributed to customers and then serviced. For example, AI is bringing automation and efficiency in marketing/advertisement collateral generation and placements. The sales process is being affected by AI automation and agents. Klarna’s customer support AI agent is already doing work equivalent to 700 human agents bringing the cost of customer support down drastically.
On the customer side as well, the way someone discovers or buys a product will change. This article by Tomasz Tunjuz on the changing enterprise buying process is a very interesting one. We have seen examples of LLM-based products like Perplexity already driving a decent chunk of traffic on some startup websites. Personal assistants (powered by AI agents can fundamentally change how businesses and individuals find and adopt a product. All the changes like these will trigger innovation in established business models and structures.
“Promise of AI is no UI”- Naval
We believe that the current UI/UX for tech products will change with AI proliferation. UI has always evolved in sync with new technology (both hardware and software) shifts. In some cases, UI elements have kept continuity during a technological shift (like the QWERTY keyboard which mimics the typewriter), or in other cases newer elements became ubiquitous with the new technology providing a different primitive (Infinite scroll coming along with touchscreen smartphones is a good example). In 2023, there was a lot of talk about chat being the dominant UI for upcoming applications (both for consumers and businesses), while that is a possibility, it is not definitive that this will be the only case. One thing is sure, the UI for technology applications will evolve with AI.
This is an exciting time for builders. AI as technology provides new primitives for entrepreneurs to reimagine businesses and drive change in the global economy. During the next few weeks, we will continue to share our thoughts across different themes/categories for Gen AI like Consumer, B2B, Developer stack, etc. We are super excited about AI and are already investors in AI-native companies like SuperAGI, Fondant, Hippo Video, and Supernova.
If you are building or planning to build with AI, reach out to us: Gaurav Chaturvedi and Natasha Malpani Oswal.