Accenture Reveals PDFs Are Major AI Token Consumers
A recording that’s been leaked from Accenture sheds light on comments from Justice Kwak, who is spearheading their agentic AI strategy. He pointed out that the heavy use of tokens in AI systems is actually coming from non-engineers, which was a surprise. Stuart Henderson, who heads the client group at Accenture, joked about how much token usage comes from turning PDFs into images and then into markdown files. Kwak confirmed that their internal data backs this up. This discussion highlights how inefficient PDFs are for AI processing due to their demanding conversion process. This was first reported by 404 Media on June 24th and later mentioned in Simon Willison's blog on August 7th, 2026, discussing businesses' worries about AI expenses.
Key facts
- Accenture's agentic AI strategy lead Justice Kwak said non-engineers drive token consumption.
- Stuart Henderson, Accenture's client group lead, joked about converting PDFs to images and markdown.
- Kwak confirmed that PDF-to-markdown conversion is a major token consumer.
- The anecdote was reported by 404 Media on June 24th.
- Simon Willison posted the link on his blog on August 7th, 2026.
- The post is titled 'The Tokenpocalypse Is Here: Companies Are Scrambling To Stop Spending So Much on AI'.
- The leak came from recorded meeting audio.
- Willison offers a paid email digest for $10/month.
Entities
Institutions
- Accenture
- 404 Media