Six months in, they tore it up and started over. 🔨 Most teams would've doubled down. Instead, the TSAI team made a breaking change and rebuilt their protocol after industry adoption shifted faster than expected. Alexander Günsche explains why on the Smooth Scaling Podcast.

Not every AI agent deserves a yes. Not every AI agent deserves a no. The future is everything in between. ⚖️ Lower limits. Higher trust. Graduated access. That's a better model than block or allow.

In this clip from the Smooth Scaling Podcast, Alexander Günsche (AWS, TSAI) starts with an unusual pizza order and ends up unpacking a bigger challenge in digital commerce. 🍕 Knowing who someone is and knowing whether to trust them are two different problems.

"Trust is more than identity." That's Alexander Guensche, AWS Senior Solutions Architect and lead of TSAI (Trust Signals for Agentic Interactions). His answer to what replaces identity: reputation. 🧵

Big news! THL Partners has entered into an agreement to acquire a majority stake in Queue-it from GRO. We're excited for what's next as we continue building the future of online traffic orchestration and helping organizations deliver trusted online experiences.

"Every trace is unique and LLMs hallucinate." 🧠 On the Smooth Scaling Podcast, Suman Karumuri explains to Jose Quaresma why the old sampling playbook is over: you can’t toss most traces anymore, and million-token contexts are inflating logs. #Observability #AI

your observability stack assumes a human is watching the dashboards. increasingly, nobody is. agents hit the APIs, don't sleep, and chase the errors you never had time for.

the most advanced way to ship software in 2026 is to put an engineer on a plane. no pipeline reaches an air-gapped cluster. someone installs it by hand on the other side.

anyone can ship a little productivity app this year. almost none of them survive the moment someone tries to charge for them. that's not a code problem. it's a scale problem.

you signed the data deal with your model provider. cool. the vendors helping you tune and evaluate that model are still sitting in the read path of everything it sees.