test_title

About This Blog

AI should make work more human, not merely faster

The better test for workplace AI is not whether it can complete a task, but whether it leaves a person more able to do the work only a person should do. Ross Barnes comes to that view through experience, not theory. He was global CTO of one of the largest agencies in the world, then later found out he has ADD and is autistic; his daughter has ADHD too. That diagnosis changes the product he builds at Galahad. His own system does not just remind him about appointments. It nudges him in different ways and helps put him in the right headspace, because he knows he often misses them. The aim is not a shinier calendar. It is a better fit between work and the human brain.

The safest automation may be the one Galahad refuses to build

Galahad’s most useful feature may be its willingness to say no. The company uses a proprietary Ikigai framework, drawn from the Japanese idea of finding humanity, to break workflows into what should stay human and what AI should take on. That can mean agents wired into Google Ads for marketing, a billing system for finance, or a rules engine in a regulated process. But Barnes sets a hard line with the four Hs: here, hunger, heart, and hunch. If a task needs presence, resilience, empathy, or instinct, the platform should not do it. His phrase for the team is plain: just because they could does not mean they should.

Shadow AI is a design problem before it is a discipline problem

People do not upload company material into personal ChatGPT accounts because they want to break policy; often, the approved tool is too hard or too slow. Galahad’s beta response is not simply to lock everything down. It inserts canaries into company content, then checks whether those canaries appear when ChatGPT is queried. If one shows up, the data has left the building. Barnes says this was tested with a large FMCG client, the kind of consumer goods company people might compare to Unilever, though he is clear it was not Unilever. The lesson is blunt: make the safe system easier to use than the unsafe workaround.

Your next meeting may include an agent that disagrees with you

The CEO version of Galahad sounds less like a dashboard and more like a small advisory room: Barnes spends an hour interviewing the executive, takes the job description and a week of calendar data, then lets the diagnostic framework find the gaps, including meetings the CEO should not attend, meetings that should not exist, and places where an agent should challenge the person in charge without becoming another employee who is always saying they are wrong.

Human content may become more valuable because AI needs it

The counterintuitive claim is that AI may increase demand for genuinely human material. Barnes argues that large language models do not want to feed on their own food. They need human content if they are going to produce differentiated output rather than recycled sameness. That is why the “Holy Smokes” stage demo matters: at a large European publisher, he built an agentic marketing use case live, aimed at getting the right content to the right consumer in the right voice while still staying safe for the organization. His bigger prediction is stranger still: within 18 months, digital avatars will attend meetings and act as people. If that happens, the scarce thing will be judgment.