Thib ๐ŸฆŠ (@CX_Architect) header
Thib ๐ŸฆŠ (@CX_Architect) profile picture

Building @SmartRole ยท CX Architect ยท Cutting through AI hype in customer service ยท Ex-Google ยท Newsletter โ†’ https://teapaw.org/t.co/qPjc5v284Q

Paris
Joined June 2011
Bad agents optimize for containment. Keep the customer away from a human at all costs. Good agents optimize for resolution. They hand off the second they're stuck, without making you repeat yourself.
1
8
The test is simple. Call your own support line as an angry customer with a weird edge case. If the bot loops you, you built a wall. If it gets you to the right human fast, you built a bridge. ๐ŸฆŠ
7
Klaviyo just launched AI agents that share the same customer profile across marketing and service. Wild that in 2026 we're calling "the two teams use the same data" a breakthrough. The bot was never the problem. The silos were. ๐ŸฆŠ
11
Genesys just bought Pinkfish. Salesforce bought Fin. Everyone's buying agentic AI instead of building it. Turns out the hard part was never the model. It's the years of messy support data nobody wanted to clean up. You can't acquire your way out of that. ๐ŸฆŠ
1
27
A new study found banking chatbots fail 88% of the time on the issues that actually matter. Blocked card. Frozen transaction. The exact moment a customer is panicking. We deployed bots for the easy questions and abandoned people at the hard ones. ๐ŸฆŠ
5
New Verint data: preference for human agents is rising. But 69% of those people said they'd happily use AI if it actually solved their problem. Customers don't hate automation. They hate automation that wastes their time. Build the second kind. ๐ŸฆŠ
3
Salesforce just bought Fin for $3.6B. Three weeks after it rebranded from Intercom. The autonomous CX wars aren't coming. The consolidation already started, and the indies are running out of time to stay independent. ๐ŸฆŠ
18
A team gave an agent one job. Tag 87 items. It got stuck in a loop, kept reloading context, and burned 1,400 dollars in under an hour. 16 dollars per tag.
1
6
This is the part nobody budgets for. An autonomous agent with no spend ceiling isn't a productivity tool. It's a taxi meter you forgot to turn off.
1
3
In the agentic era, FinOps isn't a finance problem. It's a CX architecture decision. Put a ceiling on every loop before you put a bot in front of a customer. ๐ŸฆŠ
3
A BMW dealership let its chatbot negotiate a buyback. The bot offered a deal so bad it lost them money. A court just made them honor it. Your AI's hallucination is a legally binding contract. Guardrails aren't a nice to have. They're the product. ๐ŸฆŠ
6
A hardware store just made its own website optional. Bunnings is letting customers search, pick and buy straight inside Google's AI. No storefront. No menus. No site visit. Your website was never the destination. It was always just the database. ๐ŸฆŠ
5
Klarna cut 700 support agents for AI. Re-contact rate jumped 25%. They hired them back. AI doesn't make the agent redundant. It makes the untrained agent redundant. ๐ŸฆŠ
7
Microsoft says the AI voice startups can't deliver "the control enterprises require." Translation: the giant is nervous enough about Sierra and Decagon to name them in a blog post. You don't punch down at people you're beating. ๐ŸฆŠ
18
Intercom's CEO just called out the entire voice AI space for running "fabricated demos" then dropped a fully live, unedited demo to prove the point. Half the agentic AI you saw this year was theater. The other half is sweating. ๐ŸฆŠ
6
Avaya is putting AI robots in airports and government offices. Physical robots doing customer service in the real world. We went from "will AI replace the chatbot?" to "will robots replace the front desk?" in 6 months flat. ๐ŸฆŠ
7
Every CX vendor announced an "autonomous AI workforce" this month. Zendesk. Intercom. Avaya. Kustomer. All of them. Nobody announced a plan for what happens to the actual workforce. Funny that. ๐ŸฆŠ
12
Fujitsu x Anthropic are building AI agents that evolve themselves. No human gate. No approval loop. Meanwhile most CX teams still debate whether AI should close a ticket. The gap is getting embarrassing. ๐ŸฆŠ
7
MakeMyTrip's AI agent went from 25,000 to 80,000 daily conversations in two quarters. 45% of usage from tier 2 cities. Over 200,000 bookings assisted. The best AI agents aren't replacing humans. They're reaching customers that humans never could. ๐ŸฆŠ
7
Zendesk just said they'll only charge when their AI actually resolves the problem. Not per seat. Not per ticket. Per outcome. Every vendor still billing per agent is sweating right now. ๐ŸฆŠ
11