I gave my Gensee Crate agent a single message. It searched flights, compared hotels, scraped my LinkedIn connections, planned 4 days of visits, drafted 10 personalized emails, and built a polished trip page. Here's exactly what happened.
We talk a lot about what AI agents can do. I want to show you what mine actually did.
No cherry-picking. No editing. One prompt, one result. Here's the message I sent to my Gensee Crate agent:
I'm planning a trip to Seattle in three weeks. Look for flight tickets from San Diego and hotels in Seattle. My budget is $200 per night for lodging. No set budget for flight tickets, just find the cheapest direct flights that leave on Wed afternoon and come back on Sunday morning. Also search my linkedin 1st-degree connections and find all those who live in Seattle; I'll log in to my linkedin on the browser myself, but automate every other step. Plan Thursday and Friday for two to three company/university visits where those connections work; then draft personalized email to each of them saying I would like to visit their group and can potentially give a talk on AI agent infrastructure or GPU kernel. Plan Saturday for going around Seattle; I'm interested in museum and shows. Do not book any flights, hotels, or send emails. Just give me detailed itinerary, booking links, and email drafts.
That's it. One message. No follow-ups, no clarifications, no hand-holding.
Here's what came back.
The Result
The agent produced a complete, interactive trip page — a polished dark-mode HTML document with every section you'd need. Here's the structure it generated (personal details redacted for privacy):
seattle-trip-mar2026.html — Generated by Gensee Crate
🌲 Seattle Trip — March 25–29, 2026
Company & university visits · AI agent infrastructure / GPU kernel talks · Seattle weekend
San Diego → SeattleNonstop both waysThu–Fri: 4 visits plannedSat: Museums + Show
2:37 PMDepart SAN — American Airlines nonstop5:48 PMArrive SEA — head downtown (~45 min)6:30 PMCheck in to hotel7:30 PMDinner — Canlis / Radiator Whiskey / The Pink Door
THU · MAR 26University of Washington + Microsoft🎓 + 🏢
9:30–11 AMAnon. A — Professor, UW Bespoke Silicon Group11 AM–12 PMAnon. B — Endowed Professor, UW12:00 PMLunch on/near UW campus2:00–3:30 PMAnon. C — Researcher, Microsoft Research Redmond3:30–5:00 PMAnon. D — Technical Fellow & CVP, Microsoft Azure
FRI · MAR 27NVIDIA + Anthropic🏢 Company Visits
9:30–11:30Anon. E — Sr. Distinguished Engineer, NVIDIA11:30 AMLunch1:30–3:00 PMAnon. F — Member of Technical Staff, Anthropic3:00 PMOptional: Anon. G (Google) or Anon. H (Oumi AI, Kirkland)7:00 PMDinner — RN74 / Lola / Toulouse Petit
SAT · MAR 28Seattle Exploration🌆 Leisure
8:30 AMBreakfast at Pike Place — Beecher's Cheese / Piroshky Piroshky9–10:30 AMPike Place Market — fish throwing, vendors, original Starbucks11 AM–1:30Seattle Art Museum (SAM)2–5 PMMuseum of Pop Culture (MoPOP) — Frank Gehry building5:00 PMSpace Needle (optional)7:00 PMShow — Seattle Symphony / PNW Ballet / Showbox / Moisture Festival
SUN · MAR 29Return✈️ Transit
7:00 AMCheck out9:00 AMDepart SEA — American Airlines nonstop11:48 AMArrive SAN ✅
👥 LinkedIn Connections — 20 found in Seattle area
Name
Role
Company
Anon. A ⭐
Professor & Director
UW
Anon. B ⭐
Endowed Professor
UW
Anon. C ⭐
Systems Researcher
MSR Redmond
Anon. D ⭐
Technical Fellow & CVP
Microsoft Azure
Anon. E ⭐
Sr. Distinguished Engineer
NVIDIA
Anon. F ⭐
Member of Technical Staff
Anthropic
Anon. G
SWE, AI Infra
Google
Anon. H
VP & Distinguished Scientist
AWS
Anon. I
Engineering Manager
Databricks
Anon. J
CEO
Oumi AI
+ 10 more connections found (Databricks, Google, Oracle, Meta, Shopify, UW, etc.)
✉️ 10 Personalized Email Drafts
A
Anon. A ⭐
Professor · UW Bespoke Silicon Group · Thu AM
Primary · GPU Kernel
Subject:Visiting Seattle Mar 26 — Talk on GPU Kernels + Visit to [Group Name]?
Hi [Name],
I hope you're doing well! I'll be in Seattle the week of March 25 and would love to stop by UW and reconnect. The work coming out of [your group] aligns closely with what I've been thinking about on the GPU kernel and custom hardware side.
I'd be happy to give a talk to your group on AI agent infrastructure and GPU kernel optimization — covering how modern agent systems stress standard GPU scheduling and memory hierarchy assumptions.
Would Thursday March 26 work for a visit?
Best,
[Your name]
E
Anon. E ⭐
Sr. Distinguished Engineer · NVIDIA · Fri AM
Primary · GPU Kernel
Subject:Visiting Seattle Mar 27 — Talk on GPU Kernels?
Hi [Name],
Hope you're doing well! I'll be in Seattle the week of March 25 and would love to find time to meet up. We share quite a few mutual connections and I've long followed the compiler and kernel work coming out of your team.
I've been spending a lot of time on GPU kernel design for AI agent workloads — particularly how inference patterns like variable-length sequences, sparse attention, and concurrent multi-model serving break standard kernel assumptions. I'd be happy to give a talk to your team.
Would Friday March 27 morning work?
Best,
[Your name]
+ 8 more emails (Anthropic, Microsoft Research, Microsoft Azure, AWS, Databricks, Google, Oumi AI) — each personalized to the recipient's role, company, and research area
💰 Budget Summary
✈️ FLIGHTS
~$537
🏨 HOTEL (4N)
~$600
🍽️ FOOD+TRANSPORT
~$350
🎭 ACTIVITIES
~$140
Total Estimated~$1,600–1,700
Names anonymized for privacy. The actual output included full names, LinkedIn profile URLs, and complete email drafts for each contact.
What It Did — Autonomously
Let's break down what the agent actually accomplished from that single message:
Completed autonomously:
Searched flights on KAYAK — found the cheapest nonstop afternoon departure ($537 round-trip) and morning return
Compared 6 hotels within the $200/night budget, with star ratings, price ranges, and booking links
Scraped my LinkedIn for 1st-degree connections in Seattle (I logged in myself; the agent automated everything else) — found 20 connections across major tech companies, research labs, and universities
Planned 4 days of visits — grouped by geography (university + nearby tech campus on Thursday, two company visits on Friday), with travel times between locations accounted for
Drafted 10 personalized emails — each one tailored to the recipient's role, company, and research area, with specific talk topics and proposed dates
Planned a Saturday itinerary — Pike Place Market, Seattle Art Museum, Museum of Pop Culture, Space Needle, and 4 evening show options with links
Added dinner recommendations for each evening, restaurant suggestions matched to the neighborhood I'd be in
Built a budget summary — flights + hotel + food + activities = ~$1,600–1,700 total
Generated a polished HTML page with dark mode design, booking links, and a quick-links section
And crucially: it didn't book anything or send any emails, exactly as I instructed. It respected the boundaries.
What Made This Possible
This task required capabilities that most cloud OpenClaw hosts simply don't offer:
Browser access — the agent needed to visit KAYAK, Expedia, LinkedIn, and multiple other sites. It opened a real browser, navigated pages, clicked through search results, and extracted structured data. On most hosting services, your agent can't visit websites at all.
Human-in-the-loop browser — for LinkedIn, I logged in myself (typing my credentials in the browser window, which Gensee Crate never saves or accesses). Then the agent took over and automated the scraping. This hybrid approach keeps sensitive credentials private while letting the agent do the heavy lifting.
Gensee Search — for flight prices, hotel availability, restaurant recommendations, and event listings, the agent used our #1 Source Bench-ranked search to find high-quality, up-to-date information.
Frontier AI model — the planning, email drafting, and itinerary optimization required strong reasoning. Gensee Crate serves Qwen 3.5-397B on our own NVIDIA GPUs — a model that competes with GPT-5.2 and Claude on benchmarks.
💡 "I'll log in myself" — the key design insight
Notice that I said "I'll log in to my linkedin on the browser myself, but automate every other step." This is exactly how Gensee Crate is designed. Your agent has a browser you can see and interact with. For sensitive logins, you type your credentials directly — they're never stored, never shared with the AI model. For everything else, the agent handles it. It's like screen-sharing with a very capable assistant.
A Simpler Example: Updating a Resume
Not every task needs to be a multi-day trip plan. Here's a simpler one — I asked my agent to summarize my 2025 resume, research what I've done since then, generate an updated CV, and download it:
Step 1 — Summarize & Research: The agent reads my 2025 resume, then searches the web to find what I've done since — new publications, talks, projects.
Step 2 — Generate & Download: The agent creates my updated CV incorporating the new information, converts it to a clean HTML page, and makes it available for download.
Two videos, one continuous task. The agent read my old resume, found my recent publications and projects online, wrote an updated CV, formatted it as HTML, and packaged it for download — all from one prompt.
Try It Yourself
Both of these tasks — the Seattle trip and the CV update — were done on Gensee Crate's free beta. Same models, same browser access, same tools you get when you sign up today.
