CTO @DiaIectica · Built @OrfiumMusic from 0→700 people across 7 countries · PhD, 25 publications in APIs & Semantic Web · Writing on Agentic Web & 20x teams
For engineering leaders scaling teams:
The bottleneck is no longer coding speed. It’s orchestration and taste.
Your highest-leverage move right now isn’t another headcount. It’s an Agent Operations function that turns these new plugins, workflows, and custom tunings into repeatable systems.
Stop bolting AI onto legacy processes. Start architecting the new operating system for your org.
Automate the trivial.
Explore the infinite.
Big Tech just moved in 24 hours.
Microsoft at Build: 7 new MAI models. MAI-Thinking-1 (first clean in-house reasoning engine, no third-party distillation) + MAIA 200 silicon delivering 30% better perf/$ than GB200. Full-stack control + Frontier Tuning for your own agents.
OpenAI: Codex now ships role-specific plugins (sales, analytics, design, banking) + Sites for instant shareable apps. Agents that act like specialists, not tools.
Trump EO adds runway for deployment with security review.
This isn’t model hype. It’s the architecture layer finally shipping.
First principles: Intelligence is commoditizing. The durable advantage is the integrated system — silicon + model + workflow + human judgment at scale.
Microsoft owns the stack. OpenAI is making agents portable and role-native. Policy is removing friction on power, procurement, and cyber.
The orgs that win won’t chase the next benchmark. They’ll redesign around human architects + agent swarms that compound output 10-100x.
Everything else is noise.
@kimmonismus That shouldn't come as a surprise.
I remember a year ago, the CEO of @Microsoft mentioning that the most valuable apps of the future will be the "mailboxes or hubs" of agents.
And now, a year later, they release it.
Announced today at #MSBuild: Microsoft unveiled Majorana 2, a next-generation topological quantum chip developed with the help of Microsoft Discovery’s agentic AI. https://t.co/esVcmeWdgh
@merts_dev@ManusAI Some, but the integrations are not as rich yet.
And some custom MCPs I tried to connect didn't work.
For the ones though, out of the box, the experience is even better than @claudeai.
@danleedesk Policy catching up to tech velocity. Good. But leadership test is turning regulatory tailwinds into defensible systems, not just faster capex.
@Pirat_Nation Classic picks-and-shovels.
Trillions in capex before unit economics stabilize—see railroads, internet.
The winners won't be those burning fastest but those who architect systems that compound value faster than spend.
Perplexity Search as Code is the right move.
Model writes Python to compose async search plans (fan-out, dedupe, rank) instead of serial tool-call loops that bloat context and stack latency.
This is how you sudo scale agents.
Architected, not assembled.
Production still needs sandboxes + observability on the generated code.
Introducing Search as Code, our new search architecture for AI agents.
It writes Python that calls our search stack directly, instead of looping through function calls one at a time.
Available in the Perplexity Agent API, and now default in Computer.
https://t.co/ut6GGWQTVO
@SouthernEquity@TheStartupsMag From execution to orchestration. Finally. But most will fail at the governance layer.
One poisoned agent and boom.
Scale responsibly or watch the 60x become a liability.
@EvanKirstel CEOs owning strategy is table stakes.
The gap is governance and flywheels.
Most treat AI as pilots instead of turning the wheel: embed agents in real workflows with human oversight loops.
It would be awesome to have an "auto" mode in @ManusAI that automatically selects the agent mode based on the task instead of me having to size it every time.