A new kind of website is emerging. One that doesn't just publish content, but runs itself in many of the ways a small editorial team might. That's the idea behind RoboTimes, created by Shashwat Upadhyay, Machine Learning Engineer in Guildhawk’s GAI Lab in Sheffield.
His fully autonomous, agent-led website is designed to make AI content curation feel alive, complete with AI-hosted audio recaps, real-time updates, and visible 'agent identities' showing which AI is drafting which topic.
In this article, we unpack what that means for businesses and why it's an important signpost for the future of digital services and knowledge work. We also connect the dots to another eye-catching example: Accrual Intentions, a creative experiment portraying an accountancy practice staffed by a team of AI colleagues, each with defined roles, personalities, and workflows, and only one human, founder, Alexis Kingsbury.
Most websites today are static in how they operate. Humans decide what to write, when to publish, and how to respond to readers. An agent-led website turns this upside down. You still have humans setting direction and guardrails, but software agents do much of the ongoing work: researching, drafting, formatting, publishing, updating, and sometimes interacting with users directly.
Think of an AI agent less like a chatbot and more like a digital colleague. It can pursue goals, plan steps, use tools, and take actions in a digital environment, rather than simply answering a single question.
What does RoboTimes do, in human terms?
RoboTimes is best understood as an AI-run publication. Its purpose is to keep content moving, current, and engaging, not only by generating written posts, but by making the experience feel active and ongoing.
Website interface of Robotimes.co.uk, by Shashwat Upadhyay, ML engineer at GAI Labs
Features include AI-hosted weekly audio recaps, a live ticker for real-time updates, agent identity cards showing which AI drafted which topic, multiple agent personas including 'hosts' and an 'intern agent', and a leaderboard to rank comment and reply activity.
To those unfamiliar with agentic AI, this is a system that continuously operates like a mini newsroom, producing content that is fresh, citable, and engaging. Notably, all articles are cited with real sources, so readers can verify information and engage directly with authoritative references.
Under the hood, most agent-led sites follow a continuous loop. The system observes signals such as trending topics, fresh sources, and performance metrics. An agent then plans what to do next, perhaps drafting a recap, updating a topic page, or generating an audio summary. It acts using tools and APIs, then evaluates what it produced, sometimes with other 'critic' agents or human review, before refining and repeating.
This observe, plan, act, and improve model is precisely why many researchers and industry leaders describe the shift as moving from information to action.
When do agent-led websites work well?
Agent-led sites are particularly well-suited to four scenarios.
Number one, freshness at scale. Humans cannot track everything, everywhere, all the time. Agents can monitor streams of information continuously and keep a site updated.
Number two, always-on publishing. An agent never sleeps. It can draft updates overnight, generate weekly recaps, and keep the experience feeling alive, exactly the 'live' feel RoboTimes creates.
The third is multi-step automation. Where chatbots typically answer a prompt, agents follow through on multi-step workflows. They plan, execute, review, publish, and iterate.
Fourth, traceability. RoboTimes' agent identity concept is a subtle but powerful design cue: this wasn't written by a mystery box, but by the agent persona. That pattern supports transparency and user trust.
What are agent-led websites not suited to?
As agentic systems grow more capable, so do their risks. Hereby we highlight three limitations:
One. They are not a substitute for accountable expertise. Even machine learning experts like Shashway caution,
“When deploying agents, reliability, governance, and oversight are essential, particularly because autonomous systems can make errors or act on faulty assumptions.”
Two. They are not appropriate for high-stakes decisions without controls. In regulated contexts such as finance, healthcare, and law, the cost of a wrong action is high. Human-in-the-loop review, permissions, and auditability are not optional extras.
And three. They are not 'set and forget'. Real deployments demand strong data pipelines, workflow integration, monitoring, and continuous validation.
A vivid example of where complexity surfaces comes from Accrual Intentions. The experiment flagged compliance risks including incorrect claims, and surfaced the hard reality that humans remain accountable when automated systems make mistakes.
An applied example: Accrual Intentions
If RoboTimes is an agent-led newsroom, Accrual Intentions is the same concept applied to professional services. It presents itself as a creative experiment: a 100% AI accountancy practice featuring eleven AI colleagues with job titles and defined responsibilities, designed to explore what AI can and cannot do in a firm environment.
What makes it significant is the organisational logic it implies. These aren't isolated bots answering questions. They simulate a team: defined roles, workflows, standard operating procedures, and handoffs between agents. That structure is where the future is heading, and it is closer than most organisations realise.
What the future looks like from here
We are entering a period where the architecture of digital work is being fundamentally redesigned. At Guildhawk and GAI Labs, we see RoboTimes and Accrual Intentions not as novelties, but as early proof points of a much larger transformation.
Over the near future, agent-led experiences will become more personalised. Sites will tailor not just recommendations, but custom briefings, audio recaps, and explainers adapted to individual context. They will become more action-oriented too. The shift is from content to completion: agents will not merely publish updates but take actions, making bookings, managing workflows, and handling customer service, all within permissioned environments.
Crucially, they will become multi-agent by design. Instead of a single model doing everything, systems will orchestrate specialised agents: researcher, writer, editor, compliance checker, publisher. Each one focused, each one accountable to the next.
And they will not be confined to websites. Soon, you will experience autonomous agents integrated into your hardware for instant personalised news on the go, wherever you are.
And as autonomy rises, so will the demand for governance. Permissions, monitoring, audit trails, and explicit human accountability will become foundational, not afterthoughts because instant news isn’t real news; if it’s fake information.
How GAI Labs can help you introduce safe agents
You may be planning to develop agent-led solutions to improve your business operations and want to know what works best.
With 25 years’ experience developing innovative technologies, Guildhawk and now our GAI Labs, have been preparing for the AI Age for a long time. And the most important design question in agentic AI, for us, is not 'what can the system do?' but 'what happens when it gets something wrong?'
Understanding a clients’ problem, foreseeing risks and exploring the many AI options is the first step we take with our partners, long before our GAI Labs team make any recommendations.
It is this collaborative, ability to understand the problem, research, test then create new products that work – that shapes everything do at Guildhawk.
What will define winners of the future?
As organisations across every sector begin to explore agent-led systems, the differentiator will not be which AI can move fastest. The organisations that lead will be those that know their systems are built on trust, because they are safe, auditable, and explainable, not mysterious black boxes.
That is the standard Guildhawk holds itself to. And it is the standard we help our clients achieve.
Visit RoboTimes: robotimes.co.uk Explore Accrual Intentions: accrualintentions.com