It's day one of 2025. Here's what I think will happen over the coming year.
As real-time conversational AI continues to improve (e.g. in latency, emotional expressiveness, etc), we'll see a dramatic shift from typing to speaking.
Typing will increasingly feel like a chore. Products that alleviate the burden (think @WisprFlow and @aquavoice_ ) will see significant uptake.
A new wave of voice-only apps will emerge, capitalising on advancements in conversational AI and the growing ubiquity of always-on, wireless devices like AirPods and Meta's smart glasses.
Many traditional software products will begin baking conversational capabilities in. Utilitarian apps like WhatsApp will push voice features further into the spotlight.
The transcription of virtual meetings will become the norm. Tools like @meetgranola, which make capturing conversational data effortless, will gain massive adoption.
As transcription becomes ubiquitous, we'll recognise that conversations generate vastly more data than written communication (46 quadrillion words per year vs tens of trillions according to Perplexity). The ability to harness this unstructured data and integrate it with structured data will unlock valuable new insights for organisations.
As individuals and teams become accustomed to transcribing their virtual meetings, they'll increasingly expect the same for in-person meetings. Traditional pen-and-paper note-taking will feel inadequate, leaving critical context on the cutting room floor.
Corporate transcription policies and etiquette will become key topics of debate.
Most AI products will evolve to handle two types of queries: simple and complex (aka fast and slow). Simple queries will be near-instant, while complex ones - requiring multi-step reasoning and/or external data retrieval - will take longer. Users will need to choose between speed and depth. Models like o1 and o3, tools like Google Gemini's Deep Research and features like @perplexity_ai Reasoning mode exemplify this trend.
Hour-plus response times will become commonplace.
Highly complex, reasoning-heavy queries will make thoughtful prompt design essential. The cost of a poorly crafted prompt will extend beyond money to time, as complex model chains and slower response times reduce the margin for error.
As we begin running multiple complex (slow) queries in parallel, the need for task management systems will become key. These systems will centralise tasks across various tools and will likely include features like priority queues and progress monitoring.
Agents will master interacting with systems (e.g. browsers, apps, etc) in the same way humans do - through user interfaces. This will enable them to begin tackling the long tail of digital tasks that were previously too niche or too complex for automation.
Knowledge work will gradually shift from task execution to orchestration. Rather than performing tasks directly, knowledge workers will focus on quality assurance, providing input, and delegating work.
As more day-to-day work shifts to managing and overseeing active agents, centralised notification systems will become essential for alerting us when tasks require human input, are completed, or have failed.
We'll see a wave of headless agent products - those without a frontend - designed solely to be called upon.
The "always on" era of AI will arrive, with screen-sharing assistants gaining widespread adoption. These companions will actively observe and assist, monitoring our actions and offering suggestions and insights.
Privacy and resource management will become critical challenges. These companions will require systems to determine when to observe, when to assist, and when to stay quiet. We'll see the emergence of AI companion "modes" - ranging from ultra-attentive to minimal oversight. On-device LLMs will play a key role in making these companions both feasible and privacy-preserving. The winning products in this space will master the balance between presence and interruption.
Apple and Google will likely dominate here over time.
Tens of millions of people will build their first, fully functioning software product as fully autonomous software development tools (e.g. @lovable_dev, @stackblitz, etc) become highly effective end-to-end.
We'll see many one-person 'app shops' emerge, generating significant revenues. We'll also see a wave of 5-20 person app factories.
Chat interfaces for these mass-market software products will be complemented by canvas interfaces (think @tldraw), as the realisation grows that most people need to visualise their ideas vs writing them out.
Contrary to what some may expect, more people *will* actually learn to code. In a world where short-form videos dominate and attention spans shrink, the traditional "hello world" tutorial will be replaced by dynamic, personalised app-building experiences. Imagine @cursor_ai not only generating your app but guiding you with real-time, simplified voice instructions that walk you through every step.
More capital will inevitably shift towards deep tech as software becomes brutally competitive.
As non-engineers move beyond the constraints of platforms like Wix and Squarespace, an endless stream of unique, creative websites and apps will populate the web, bringing us back to an era of individuality on the internet. True self-expression through pixels will no longer be reserved for those who can write code.
Exploring the internet will once again feel fun, as conversations are had and new relationships are built on the edges, not only in digital town squares.
Some apps and websites will be intentionally designed with shelf-lives in mind. As the cost of building software plummets, ephemeral applications will become a thing.
The ongoing democratisation of software development will make design more important than ever as a means of standing out. Unlike good code, good innovative design will remain harder to model and automate (at least for a while longer).
Micro interactions will increasingly determine the winners and losers in highly competitive product categories.
X will continue to play a central role in global politics and the dissemination of real-time news. We'll also see the emergence of new challenger platforms focused on citizen journalism and open source intelligence.
A growing number of new-age news businesses will emerge to challenge incumbent publishers, founded by solo creators or small, decentralised teams. These businesses will distribute content exclusively on social platforms, leverage AI for both research and content creation, and monetise through subscriptions and donations
AI-generated content will become increasingly difficult to detect without some form of authentication. As the incentive to spread disinformation for engagement continues to grow, AI-powered disinformation campaigns will become rampant, making them an increasingly pressing problem.
Community notes will continue to define the standard for fact-checking on large social platforms.
Following the 2024 Presidential Elections, podcasts will continue to be the undisputed medium for deep, engaging debates.
TikTok gets banned?
Algorithmic feeds and the endless stream of content, made easier than ever by AI, will drive us to peak information overload. This will create a demand for a new category of personal agents dedicated to curating and summarising content for us. These agents will understand what matters to us and when, filtering out irrelevant information, distilling narratives into bullet points, and offering differing viewpoints. They will also allow us to consume content in whatever format we prefer - text, audio, or video.
The majority of the content we consume daily will be fully AI-generated or AI-enhanced (it probably already is).
It will become easier to train models to adhere to specific writing styles. Those with large archives of personal written content will have a distinct advantage in exponentially increasing written output while maintaining their true tone of voice.
Phone addiction will continue to grow into a massive societal problem.
Seamless dubbing (thanks to companies like @elevenlabsio) will make content globally accessible from day one, significantly expanding creator reach and monetisation potential. This will also mean that competition among online creators becomes cut-throat.
Increased competition will continue pushing many creators to make very questionable decisions to drive views.
The average online creator will struggle to sustain their audience's attention as AI makes it easier than ever for new creators to compete with high-quality, engaging content. As a result, creators will continue pushing to own their audience, which will serve as a net tailwind for platforms like @circleapp and @SubstackInc.
We'll be inundated by an endless stream of emails and messages from AI sales agents, AI recruiter agents, etc. The constant flow of AI-generated content flooding our inboxes will create a need for ambient inbox managers. If 2024 was the year for companies building outbound tools (think @11x_official), 2025 will be the year they build for inbound.
Messages and emails we receive will be summarised by AI, while those we send will be automatically tailored and simplified for recipients. This shift will diminish the role of traditional prose and formalities, making interactions more efficient but less personal. Email will adapt to this new reality (think Cora by @every) or evolve into an entirely new AI-native protocol. Communication will increasingly follow a human-to-agent-to-agent-to-human flow.
Continued advancements in AI-generated video will force the major studios to confront the reality that their billions in physical assets could become obsolete within a few years, as AI models replace cameras and sets.
A wave of challenger AI studios and agencies will emerge.
AI video will begin to dominate the advertising and music industries. Low-quality video ads on social media will fade away, replaced by far more engaging AI-generated content. Meanwhile, previously under-the-radar artists from Spotify's long tail will go viral with the help of AI-generated music videos and social. AI video will also start to seep into the film industry, hinting at what's to come.
Producing high-quality AI video will remain a true craft.
The fragmentation of models - spanning both closed-source and open-source across modalities like video, image, and audio - will keep the process of video production inherently complex. Achieving a cohesive result will require chaining together multiple models, demanding both technical expertise and artistic vision.
A surge of interest in AI video production will bring many new creators into the field.
Demand for quality video will skyrocket.
Models across all modalities will continue to improve, become more affordable (see @deepseek_ai V3), and increasingly accessible, driving greater daily use among society's long tail. However, most people will still struggle to grasp their full potential, creating significant opportunities for education across all age groups.
ChatGPT and Claude will continue growing quickly but Google won't be going anywhere. The company will remain the dominant player in search and compete aggressively. Apple will release substantial updates to Apple Intelligence, which will (hopefully) make it much better.
Copyright in AI will remain a mess.
We may or may not see AI's IoT moment with on-device models embedded in everyday appliances (e.g. fridges, ovens).
We'll see some cool experimentation around AI social networks.
AI agents will join our group chats as mods.
AI-generated nostalgia will be the defining vibe of 2025, as a swarm of creators will emerge to recreate past eras.
There will be a growing desire for in-person, device-free gatherings.
Generalists with a strong understanding of AI will become highly sought-after hires.
Traditional educational institutions will continue to lose their footing, especially as advancements in AI render much of their curricula outdated. They will struggle to keep pace and justify their fees.
AR and VR will continue to advance in the background.