AI Agents: The new digital middlemen
AI agents capable of complex, self-directed tasks are becoming a reality, with capabilities set to improve dramatically through this year, and diffuse widely.
Consumer agent uptake will be hard to time, but fast when it occurs. Enterprise adoption will happen slower but with greater inevitability, as agents offer strong productivity gains across many business functions.
TMT firms should be able to capitalise on much of these potential cost savings, but are exposed to a number of specific risks around agents acting as new digital middlemen, disintermediating traditional web ecosystems within advertising and ecommerce.
Related reports
AI’s defining year: Capex and capabilities surge
20 February 2025US big tech companies are deploying hundreds of billions of dollars to remake the global economy in their image, as enviable growth contrasts with layoffs and low morale.
The cost of using AI models will fall in 2025 and make more AI applications possible. Regulation is caught between pressure from Trump and investigations that must go on, such as digital markets.
Microsoft and Google have tied their fortunes to AI. Amazon and Meta stand to realise business gains from AI, while Apple is the outlier: capex declined in 2024 as it focuses on iPhone and services.
Telco network APIs: Real potential but still nascent
20 February 2025Telcos are increasingly developing APIs to share selected network data with third parties, with the goal of supporting useful end-user applications.
Capabilities are still nascent, but the potential is real. Telcos need to adopt a pragmatic approach that looks to match API capabilities to useful products, and build increasing scale over time.
Security is the largest near-term opportunity for API products, but AI is the key emerging area, with telcos potentially able to play an ambitious role in providing APIs to help manage the growth of autonomous AI agents.
Use of publisher content to train AI models is hotly contested. Unacknowledged scraping, licensing deals, and lawsuits all characterise the publisher-AI company relationship.
However, model training is not the whole story. More and more products rely on up-to-date access to content, and some are direct competitors to publisher offerings.
Publishers can’t depend on copyright to deliver them the value of their IP. They need to track which products are catching on with users for licensing deals to make sense for them, and to ensure their own products keep up with the competition.
Big tech meets big capex: How high can it go?
1 November 2024Big tech capex is set to jump over 50% in 2024, fueling the current AI boom, and supporting the training and deployment of the next-generation of frontier models slated for release over the next 2-4 months
If these frontier models can deliver greater capabilities, and the returns to match, it will intensify the race to scale up capex even further to train ever more powerful models on ever larger clusters of chips
If returns do not flow to the frontier, then models become commoditised, with all of big tech able to capitalise on their application layer dominance. If they do, then outcomes are uneven and uncertain with the core cloud players racing for dominance and leaving the others behind
AI’s defining year: Capex and capabilities surge
20 February 2025US big tech companies are deploying hundreds of billions of dollars to remake the global economy in their image, as enviable growth contrasts with layoffs and low morale.
The cost of using AI models will fall in 2025 and make more AI applications possible. Regulation is caught between pressure from Trump and investigations that must go on, such as digital markets.
Microsoft and Google have tied their fortunes to AI. Amazon and Meta stand to realise business gains from AI, while Apple is the outlier: capex declined in 2024 as it focuses on iPhone and services.Telco network APIs: Real potential but still nascent
20 February 2025Telcos are increasingly developing APIs to share selected network data with third parties, with the goal of supporting useful end-user applications.
Capabilities are still nascent, but the potential is real. Telcos need to adopt a pragmatic approach that looks to match API capabilities to useful products, and build increasing scale over time.
Security is the largest near-term opportunity for API products, but AI is the key emerging area, with telcos potentially able to play an ambitious role in providing APIs to help manage the growth of autonomous AI agents.
Use of publisher content to train AI models is hotly contested. Unacknowledged scraping, licensing deals, and lawsuits all characterise the publisher-AI company relationship.
However, model training is not the whole story. More and more products rely on up-to-date access to content, and some are direct competitors to publisher offerings.
Publishers can’t depend on copyright to deliver them the value of their IP. They need to track which products are catching on with users for licensing deals to make sense for them, and to ensure their own products keep up with the competition.
Big tech meets big capex: How high can it go?
1 November 2024Big tech capex is set to jump over 50% in 2024, fueling the current AI boom, and supporting the training and deployment of the next-generation of frontier models slated for release over the next 2-4 months
If these frontier models can deliver greater capabilities, and the returns to match, it will intensify the race to scale up capex even further to train ever more powerful models on ever larger clusters of chips
If returns do not flow to the frontier, then models become commoditised, with all of big tech able to capitalise on their application layer dominance. If they do, then outcomes are uneven and uncertain with the core cloud players racing for dominance and leaving the others behind