Impact of AI: Explored podcast, with James O’Regan and Gerjon Kunst · December 16, 2025 · ~50-minute year-end review and 2026 predictions episode.
Brian’s third appearance on the show, now in his role as VP & Futurist at Citrix. Computer-using agent benchmarks have just crossed human parity, AI browsers have emerged as a real security concern, and MCP has gone ubiquitous. Brian is also, for the first time on the show, describing an early version of what would become his second-brain system, built on markdown files.
What a futurist actually does
The role isn’t that different from twenty years as an industry analyst, journalist, or writer — look at what’s happening in the industry, summarize it, explain why it matters, what it means, what different companies are doing. The only pivot is timeframe: instead of today, it’s one to five years out. As a Citrix futurist, Brian doesn’t work in product strategy; he paints a picture of what the world looks like in one to five years and shares it with colleagues so they can build strategy, then goes on stage and tells those stories directly — including a recent joint session with Microsoft’s own futurist, Marco Kasalena, at Microsoft Ignite: “two futurists walk into a bar and just talk about the future of work.”
What last year’s predictions got wrong
At EUC Forum in London, Brian and the hosts graded their 2024 predictions and gave themselves a B+. The biggest miss: neither of them saw AI browsers coming, or the scale of their impact. AI browsers turned out to be great for productivity and an absolute security nightmare in the same breath.
Shipping in public, and who pays for the mistakes
AI labs are running their proofs of concept in public in a way no other technology wave has — “amazing AI, ask us anything, answers may be wrong,” shipped by trillion-dollar companies. A consulting firm that shipped a $100,000 report with the same disclaimer wouldn’t survive it; some consulting firms did ship AI-assisted reports with errors, and clients responded by demanding fee reductions for work AI did instead of the humans they were paying for.
What “more productive” actually means
Knowledge workers are paid for their brains, not for physical output. If AI cuts a two-hour task to one hour, what happens to the freed hour — double the output, do the same task twice as well, or go home? A Microsoft Ignite survey asked exactly that, and the top answer was reclaiming personal time — the basis for Brian’s blog post “IT admits workers are in charge of AI. Workers admit they use it to go home at 5.” Companies expecting 20% more output from more-productive workers are misreading the moment; workers experiencing it are mostly just relieved not to be working all night anymore.
The year of agentic AI — or its dawn?
By the pure numbers, computer-using agents crossed human parity: OSWorld’s 369 tasks put the median human at 72% completion, and the best AI has gone from about 40% a year earlier to 76% today, open source and independently verifiable. That doesn’t mean every job is replaced — a human still has to direct the agent — but “go do my expense report” is close. At the same time, most individual workers are nowhere near the frontier. At MAICON, the Marketing AI Conference in Cleveland, even the speaker room conversations were startlingly basic — most workers are still at “I use ChatGPT to help write an email,” technology that’s existed for two or three years. The frontier is moving fast; the average worker isn’t there yet.
AI browsers as a security problem
Testing Comet and Atlas, Brian found you can hand them company credentials and they’ll log into Office 365, set appointments — genuinely useful — but the same access lets them do a great deal else inside a company context. Every company needs an enterprise browser now, not necessarily a dedicated product: Chrome Enterprise Premium with extensions, Microsoft Edge for Business, anything that logs on with work credentials and enforces DLP, screenshot prevention, and copy-paste prevention. Enterprise browsers have been mature and easy to operate for five years. The era of any random browser on any random device is over.
MCP: imperfect, not broken
Microsoft alone had roughly 30 separate MCP-related announcements at Ignite. Security across MCP servers varies wildly — Andrew Morgan’s “Python honey pots” line lands — but don’t let great be the enemy of good; plenty of protocols that later became standards looked like garbage early on, and there will be an MCP 1.1 that fixes today’s gaps. Microsoft’s Work IQ concept at Ignite reframes the model entirely: instead of a connector pulling data out, an agent runs inside the secure environment itself, like a librarian who asks what you’re actually trying to do, checks every permission internally, and hands back only the results you’re allowed to see.
2026 predictions
IT departments, currently focused on AI FOMO (”don’t be the Blockbuster in the Netflix documentary”), will start waking up in 2026 to the fact that workers using AI in ways IT doesn’t understand or approve is a real business risk. A financial correction in AI markets seems likely, but a stock correction doesn’t invalidate the technology — pets.com crashed in 2000 because nobody wanted dog food by mail, not because the internet was fake, and today OpenAI has 800 million weekly users regardless of what happens to valuations. Brian’s personal prediction: he’ll be using AI more, not less, twelve months from now.
The early second brain
Every blog post gets copied into a markdown file — by this point, roughly 40,000 words of the past year’s writing, dropped into every new chat as a style reference so the model never has to guess how Brian writes. Claude 4.5 in particular nailed his voice and way of operating.
Key formulations
“I don’t think anyone’s talking to apps anymore. I think everyone in the future is talking to AI.”
“IT admits workers are in charge of AI. Workers admit they use it to go home at 5.”
“We all live in the AI bubble. Just because the technology exists at phase five doesn’t mean every human is there.”
“The days of just any old random browser from any random device — those days are over.”
“Don’t let great be the enemy of good.”
“I don’t care if valuations go from 5 trillion to 1 trillion. I still really love using ChatGPT.”