Luc Julia on voice AI: what Siri's co-creator told Lineshift
In June 2025 the Lineshift team attended a talk led by Luc Julia, co-creator of Siri and now Scientific Director of AI at Renault. Julia is one of the few people in this industry with the track record to be blunt about his own field, and he spent the hour being exactly that.
Where did AI come from?
Since the term "artificial intelligence" was first used in 1956 at Dartmouth (USA), the discipline has alternated between inflated expectations and "AI winters." He walked through the phases:
- 1970-1990: invention of expert systems and logical decision trees, the first attempts to replicate expert human reasoning.
- 1997: Deep Blue (IBM) defeats Garry Kasparov at chess.
- 2000s: machine learning takes off, thanks to the internet and big data.
- Since 2010: rise of deep learning and of generative AI.
Julia's core claim: AI is not intelligent per se; it's a highly specialized tool — faster, more available, more reliable than humans on certain tasks, but not more "intelligent." He places it in the lineage of Pascal's calculating machine, the modern calculator, and GPS.
Siri: the origin of a voice intelligence
- In 1997 Luc Julia and his team built one of the early conversational voice assistants. It only correctly interpreted about 70% of words.
- Their fix became the "nightclub theory": people in a loud nightclub get by on humour and context. That "more human" idea appealed to Steve Jobs, who acquired their company.
- At one point Siri had about 180,000 users. When it launched on the iPhone on October 4, 2011, it quickly jumped to 180 million.
- Since then Siri hasn't evolved into a full-dialogue assistant; most of its uses remain single-step commands ("Call mom," "play music," "what's the weather").
Why he calls global generative AI an ecological dead end
- Generative models like ChatGPT and Gemini are very powerful: ChatGPT-4o uses over 1.2 trillion parameters drawn from all over the internet.
- Running these models requires huge resources: data centers consume large amounts of electricity and water.
- Julia argues older models were more efficient and "frugal" — and despite the resources, today's models still make many errors, about 36% by his estimates.
- He believes the future lies in vertical, domain-specific agents rather than one general AI.
What comes after general AI?
He advocates targeted, precise agents specialized in specific domains, like Lineshift's automotive voice assistant. At Renault, he has overseen an onboard agent that is "lean and relevant" in its new R5 model.
That message is uncomfortable for anyone selling general-purpose AI and convenient for a company like ours. We think it is simply accurate: a voice agent that knows nothing about your stock, your workshop, or your DMS will not book a service appointment, whatever its parameter count. The generalist-versus-specialized trade-off for dealership voice AI is exactly this argument in practice.
He ended on career advice: learn logic and mathematics, not just programming languages, which change. And borrowing Steve Jobs's "Stay hungry, stay foolish," he warned against being passive about AI: understand it, tame it, use it to build a more human future. To hear what a specialized agent actually handles, the callbot guide for car dealerships walks through it.
If Julia's case for specialized agents describes a problem you hear every day on the phone, Contact Lineshift AI and we will show you what a vertical voice agent does for a dealership.
