J'aime bien démarrer une nouvelle année en fanfare, alors quoi de mieux que d'animer un podcast ! Comme on dit souvent, c'est tout ou rien.
In taking on this podcast as the host going forward, I wanted to take the conversations with topics like AI to a more in depth and tangible feeling place. I’m incredibly passionate about this topic and my goal is to hold deep, useful conversations about where AI is headed and what it means for the folks who try to make it work every day, which I call the Future of Work.
Pour commencer, je me suis entretenu avec deux personnes qui, à mon avis, ont des points de vue très intéressants sur l'IA : Patrick McGarry, Federal Chief Data Officer chez ServiceNow, et Dr. Jupiter Bakakeu, Lead Generative AI Technologist chez Alteryx. Ce ne sont pas de simples commentateurs. Ils sont sur le terrain, à bâtir l'avenir et à réfléchir à la façon dont l'IA fonctionne réellement, et ils ne cherchent pas à édulcorer la réalité de ce que cela exige.
To get a taste of where this amazing conversation went, take a look at these 5 ideas from this episode that I think will shape the AI conversation this year:
1. Les agents changent les enjeux
AI agents are moving away from the perception of just being smart assistants to shifting the entire landscape of how teams are formed and work is done.
Jupiter broke it down like this: Real agents don’t just generate answers; they perceive, plan, act, and learn. Once they start taking real-world actions (like modifying records or triggering workflows), the stakes go up. Fast.
I think Patrick nailed it: ”The moment AI can take action, you have to care about governance. Mistakes aren’t just wrong answers anymore. They’re real outcomes.”
This is where a lot of the hype falls apart. Cool demos don’t mean a system is safe, auditable, or even reversible.
2. Il faut déléguer avec discernement
Everyone’s excited about delegation, and for good reason, AI agents can be powerful teammates.
But just handing off everything is a sure way for disaster.
I loved Jupiter’s framework on this point: Look for tasks that are repeatable, reversible, and auditable. If it meets all three, go ahead and delegate. If not? Keep a human in the loop.
Exemples :
- Organiser vos fichiers ou classer des documents
- Planifier des réunions ou résumer des rapports
- Prendre des décisions financières
- Produire des déclarations fiscales
Bottom line: Just because an agent can do something doesn’t mean it should. We need to be thoughtful about this part!
3. Voice is rising, but it’s not the end of the UI
We talked a lot about voice. Honestly? I’m bullish on it, especially for consumer use cases.
Voice lowers friction. It makes delegation feel more natural. And when it works, it feels like magic. But enterprise use cases? I’m willing to admit, that’s a taller order.
Patrick made a great point: ”Voice may feel natural, but governance needs the receipts.” When you’re working on regulated tasks or precision-heavy workflows, clicking a button is still safer than hoping an AI caught your meaning.
Jupiter was on the same page: “Voice should augment your interface, not replace it.”
Dans les questions rapides en fin d'épisode, ils ont tous deux reconnu que la voix allait évoluer, mais que la saisie au clavier resterait la norme pour l'instant. Et moi je serai de l'autre côté en train de parler à mon ordinateur !
4. Les principaux obstacles ne sont pas d'ordre technique
One of the biggest myths in AI right now is that model quality is the bottleneck. It’s just not the case.
The real blockers are trust and cost.
We’re talking massive infrastructure needs and there’s a reason compute costs are exploding. Then layer on global regulations like the EU AI Act, amongst many others in flight, and you’ll understand what real constraints look like. AI won’t move forward with out a level of trust and a lower bill.
This makes tangible sense because guardrails matter. But it’s important to recognize that the path to scalable AI runs through modernization, governance, and data quality, not just cooler models.
Patrick summed it up wisely: ”The winners in 2026 won’t chase every new feature. They’ll be boringly compliant and quietly effective.”
Je suis d'accord.
5. La meilleure IA est invisible
One nugget of wisdom came up a few times: The most impactful AI might be the stuff we don’t even notice.
Par exemple, Jupiter a créé un agent d'arrière-plan qui nettoie et classe tout son Google Drive toutes les nuits, sans bruit ni tâches supplémentaires. Juste des tâches qui se font sans que vous ayez à y penser.
That’s the bar we should be aiming for. AI that fits into your life or workflow quietly doing what needs to be done, without creating new risks.
Vous direz peut-être que ce n'est pas assez révolutionnaire ou radical. Je ne suis pas d'accord. Si chacun progresse de 1 % chaque jour, l'impact collectif devient immense en un temps record.
We’re past the phase where AI is the shiny new toy. We’re in the phase where people are seeing the way forward and becoming strategic and deliberate about the path.
That shift from answers to action is what 2026 is all about.
Listen to the full episode here: 🎧 Alter Everything Podcast: What Does AI Look Like in 2026?
J'aimerais beaucoup connaître votre point de vue. Vous pouvez laisser un commentaire ou m'envoyer un e-mail à [email protected].