Friday’s operating update: AI has officially moved from helpful copilot to the person quietly reworking the org chart. OpenAI’s GPT-5.5 wants the whole workflow, not just the prompt; Rilian’s $17.5M defence bet says deployment is the real moat; and a fresh crop of compliance, multilingual support, and expert-shadowing startups suggests every category now ships with an automation layer by default.

The catch, naturally, is that privacy, security, and trust are still doing last-minute QA.

For twenty years, finding office space in London has taken six weeks and eleven people who don't work for you. We built the fix.

Seek is the UK's first direct-to-landlord office search platform. Every major London space on one site, real prices side by side, zero fees for tenants. No agents. No runaround.

Built by Peter Dudley and Lizzy Morrison, two former commercial agents who spent two decades inside the broken process — and decided to end it.

seek-re.com

In defence, the bottleneck isn’t more tools, it’s getting any of them deployed where it counts. Rilian has raised $17.5M to scale Caspian, an AI-native command layer that sits above existing security stacks and runs pre-trained agents inside air‑gapped, sovereign-cloud, and compliance-heavy environments. The round from 8VC, First In, Tamarack Global and others is explicitly defence-first capital, not generic AI money.

Caspian replaces analysts swiveling across dozens of consoles with agentic security command layer logic that automates detection, countermeasures, and targeting, and can ship capability updates in days. The product is aimed at what CEO Christian Schnedler frames as a procurement and manpower problem disguised as a tech gap.

The cap table and early traction show the go-to-market plan: US-Israel-GCC defence corridors, a nation-scale UAE NSOC deployment, and founders wired into US national security circles, including Nick Pompeo. The tech is interesting; the distribution looks like the real moat.

GPT-5.5 is less a model release and more a new employee type: a tireless staffer that can finally drive the computer. OpenAI introduces a new class of “real work” AI that plans multi-step tasks, hops across tools, and finishes jobs with fewer retries and tokens, while matching GPT-5.4’s latency.

Under the hood, it’s tuned for agentic coding, OS-level computer use, and long-horizon knowledge work and research. Inc’s early look promises intent-level inference: the system often figures out what’s really being asked before it’s spelled out. For teams, that shifts AI from autocomplete to project co-owner.

Rollout targets ChatGPT Plus/Pro and Codex first, aiming squarely at professional workflows. The New Stack notes GPT-5.5 signals frontier-work focus, with stronger safeguards around cyber and bio. The leverage goes to companies that redesign processes so the model owns entire workflows, not just individual prompts.