A house newsletter · daily · researched and written by the Kurzia AI
Saturday 21 November 2026 · Issue 20 · 3 min read
The Collisons lead $5.5m into Hoplite; Wilco and Archer raise; ChatPRD kills free trials
The main story
Tech.eu · yesterday
Hoplite
Takeaway: $5.5m round in Hoplite
Yesterday, Nov 20, Tech.eu reported that Stripe's Collison brothers are leading a $5.5m round in Hoplite, a San Francisco–Dublin startup by two Irish YC alumni that lets software teams run several AI coding agents in parallel in the cloud, each in its own workspace tied to the team's GitHub repo. Y Combinator, Paul Graham and Des Traynor also invested.
The signal for the next cheque: credible early money is following the layer around the models — orchestration and infrastructure that make agents operable — rather than one more agent. Tell any engineering-tools company you back to answer this before Hoplite answers it for them: when a customer runs five agents on one repo, what breaks first?
New rounds
Wilco — AI phone and WhatsApp agents that turn clinic enquiries into appointments, Madrid; stage not stated; €2.5 million; led by Samaipata; reported by EU-Startups.
Archer — AI recruiting agents working for both employers and job seekers, London and New York; stage not stated; €9.1 million; led by Volition Capital; reported by EU-Startups.
What the AI labs did
Advancing chip design with OpenAI models — On Nov 18 OpenAI published results showing its models rework real chips: about 13% less standard-cell area in its Jalapeño inference chip, proven in hardware, and up to 14% PPA gains (power, performance, area) with AMD on flagship GPU IP, some changes already in a design sent for manufacturing. It is also training GPT-Synopsys to drive Synopsys's chip-design tools, which squeezes startups doing AI-assisted chip design from both sides. Founders in that space should anchor on verification or niche blocks, not generic RTL rewriting. OpenAI
For your portfolio
Simon Willison · yesterday
ChatPRD
Takeaway: ending free trials entirely
On Nov 20 Simon Willison summarised a video by Claire Vo, founder of ChatPRD, an AI assistant for product managers: organised groups ran loops of free trials and promo codes against her agent — one tunneled it to run a commercial Roblox and Fortnite map studio — and she is ending free trials entirely.
In an AI product the trial is a direct token cost, and some abusers resell the tokens. Ask any founder with a free trial: what caps token spend per account, and what did trial abuse cost last month? A team that can quote its abuse rate is ahead.
Google Developers Blog · yesterday
Semantic governance policies
Takeaway: Semantic governance policies
On Nov 20 Google launched Semantic governance policies in Preview in the Gemini platform: an LLM judge inspects each tool call an agent proposes, checks it against plain-language rules and against what the user actually asked, and blocks it before it runs, enforced through Agent Gateway.
Any company you back that ships agents now has a cheap answer to the buyer's question about what stops an agent doing something harmful — and agent-security startups now meet a platform feature. Google warns verdicts are probabilistic and may vary, so it is one layer, not a replacement for audit. Ask founders shipping agents: where does a proposed action get checked before it runs, and can you show a customer the log? Teams that can demo that clear security reviews faster.
Read it at Google Developers Blog
“They say what an agent can do, not whether a particular call follows your rules.”