For someone who founded a tech business, I’m fairly sceptical of technology – especially when it feels more like marketing hype.
My starting point has always been, what’s the problem – and what’s the best way to solve it? If there’s a technology angle, that really interests me. But more often, it’s solved through a combination of technology, new business models, and process re-engineering. Technology is the enabler, not the driver.
So while I’ve been keeping up with developments in AI, I’ve been slower to jump on the AI bandwagon. I’ve certainly been interested in its amazing potential and less inclined to believe the doomer predictions. But everything I’ve seen and heard to date has been over hyped, oversold, or just not possible – until now.
The release of Claude Cowork on 12th January 2026 changed the game and OpenClaw’s acquisition by OpenAI shortly after, has given me a vision of the future.
But let’s step back a bit.
At the beginning of the year, I decided to try my first ‘AI project’. I’d seen lots of people promoting platforms and apps they’d built in Claude Code or OpenAI’s Codex, declaring we wouldn’t need programmers anymore (not true), or claiming they’d used AI to vibe code an app, which, while fun and an important experience for them to try the tools, wasn’t game-changing.
I don’t know about you, but I have a problem keeping track of all the relevant data, emails, slide decks, articles, reports and notes I’ve gathered over the years and the increasing amount of new information that I was being bombarded with. I couldn’t find what I needed easily – and quickly connect new ideas.
Basically, I wanted the benefit of a large language model (LLM) data set but with my context. One that could then be my primary source for my chat bot of choice. My personalised knowledgebase.
I fired up Claude and worked on a scope and plan for my project. Including a step-by-step guide on which tools to use, and how to set it up. It’s a beautiful document and you can see the final result here.
Three steps in, I realised the tools Claude suggested didn’t actually do what it promised, so I asked Claude to take a closer look, and make sure it only gave me options with current features that work. After apologising, it updated my scope and I started again.
One step involved installing my own LLM on my hard drive. Worked fine – until I tried to actually run it, and my laptop crashed. It turns out, you need powerful chips to run these things. Back to Claude with my update, and it admitted that, while it’s technically possible to install an LLM on my laptop, it would not deliver the best results. And could potentially crash.
Meanwhile, in mid-January, Anthropic (makers of Claude) released Claude Cowork – a non-technical version of their programming productivity tool Claude Code.
This sounded interesting, so I dived in with a few tasks – ones that actually deliver results. Automations and workflows that were once only possible to do in Claude Code – but I’d steered clear of because it requires knowledge of coding and terminal windows – and the only terminals I’m comfortable with involve lounges and airplanes.
I chose a task I’d been procrastinating on: my expenses. I pointed Claude Cowork at my recently scanned list of 50+ receipts and asked it to file them by month, change them all to a consistent, easily understandable file name, and prepare an expense report in Excel. It does all this, and more. A beautiful report that’s broken down by month and category of expense. Without me asking.
This was exciting. I wanted more. I configured connections to Chrome, Google Drive and Gmail, and set Claude on a task to help me plan my ‘Grown Up Gap Year’ six week driving trip around France. It navigated to web pages to check accommodation availability, sourced files from my drive and suggested ‘hidden gems’. It even produced a wonderful map of the suggested route.
Finally, something useful that’s easy for non-coders to accomplish.
I asked Claude to update my knowledgebase project to use Claude Cowork and I can now connect Readwise, the tool I use to gather and make notes on web articles, podcasts, and YouTube videos, with Granola, the tool I use for recording my meetings, and combine all that with access to reports and presentations in my Google Drive. Not only can I ask Claude Cowork to follow up a meeting referencing specific research I’ve collected, I can also ask it for feedback on how I ‘turned up’ in the meeting and if I can improve in any way.
And how does this relate to the future – and OpenClaw?
Well, OpenClaw is a wrapper or operating system for agents that lets people construct automated tasks on top of a large language model like Claude or OpenAI. It’s a competitor to Claude but it doesn’t have the built-in guardrails and permission controls that Claude does. But that should change quickly with its recent acquisition by OpenAI. Expect to see improvements to the usability of Codex and ChatGPT very soon. An interesting aspect of the acquisition is that Peter Steinberger, the creator of OpenClaw, made it a requirement that OpenClaw remain an open-source tool. This could mean OpenClaw will attract more development resources and support from beyond just OpenAI and become the standard operating system for managing agents.
And here’s why this step-change matters. With the right safeguards in place, using the right sources of information, we will no longer have apps, but a whole bunch of workflows, skills and tasks that are undertaken for us. It’s now possible to foresee a future where all the mundane and boring tasks are automated – and we can focus on the value we can bring as humans.
Imagine if a digital assistant could give a prospective student comprehensive and impartial advice on all possible study opportunities. Could this change the business model, removing agent commissions? How could we completely re-engineer the admissions process to be more transparent and effective in admitting the right student, and what retraining do we need to manage a digital co-worker?
I feel more confident in the ability of the emerging technology to deliver useful ways to answer all these questions (and more) than I did at the end of last year. And I’m willing to predict that some of these changes will be more than simply automating what we currently do but will fundamentally change how we operate.
Jason Howard founded StudyLink, the world’s leading admissions and agent network platform, in 1992. He steered that business through three profound technology transitions – from print marketing to digital marketing, on-premise to cloud hosted software, and licence fee to SaaS business model – before successfully exiting in 2024. Having managed customers and users through these transitions over 30 years, Jason is in a good position to consult and advise on how the higher education sector navigates the next wave of technology change.