"Quite often it's given me more to do" — the case that AI is adding work, not removing it

At the GamesIndustry.biz HR Summit in London on 1 October, generative AI came up repeatedly — and the room was notably less enthusiastic than the marketing. The through-line: AI is not the time-saving miracle being sold, and it arrives with data-privacy worries, employee anxiety and more HR work.

Two stories worth remembering

The shader incident. An artist used Gemini to generate shader code while a programmer was out. The code was inefficient, broke company coding standards, and violated the studio's generative AI terms. It was only spotted weeks later — and the result was a build revert. The time saved was handed straight back, plus interest.

The outsourced art. Art was sent to a freelancer who used AI. It shipped with, in one attendee's description, "more fingers than most humans have." The community noticed immediately.

That time and cost saving they thought they got wasn't there.

The structural problem

Most of them collapse, because [workflows are] driven by the tacit knowledge of individuals rather than formally documented operating procedures.

That's the least quoted and most useful line of the whole event. AI tools slot best into documented processes. Most studios run on undocumented knowledge held by specific people. Pushing AI into a workflow nobody has written down doesn't automate the work — it automates the ambiguity.

Where it's genuinely helping

• Framing emails and getting thoughts in order
• Answering repetitive HR questions (maternity, sickness leave)

Where it's creating work:

• AI-written grievance letters, around 10 a week for one legal representative — long, factually shaky, with "quite odd" legal references. HR is duty-bound to read and answer all of it properly. One formal grievance ran nine pages, of which roughly eight were filler.
• Policies going stale in weeks. "I wrote a policy two weeks ago, and then I found out about some prompt injection yesterday that means I've got to go back and rewrite some of the guidelines."

The honest summary

I haven't seen AI create anything fun. It can replicate, for sure, but I haven't seen anything fun come from AI yet.

You can make crap quicker.

Read those next to the numbers: 97% of studio leaders have adopted some form of AI, and more than half rate their pipeline integration at three or lower out of ten. Both can be true. Adoption is not the same thing as integration, and this month's evidence is that the gap is where the work — and the mistakes — are hiding.