Day Zero · Session 01 · Tue Aug 25 & Wed Aug 26, 2026
██████╗ █████╗ ██╗ ██╗ ██████╗ ██╔══██╗ ██╔══██╗ ╚██╗ ██╔╝ ██╔═████╗ ██║ ██║ ███████║ ╚████╔╝ ██║██╔██║ ██║ ██║ ██╔══██║ ╚██╔╝ ████╔╝██║ ██████╔╝ ██║ ██║ ██║ ╚██████╔╝ ╚═════╝ ╚═╝ ╚═╝ ╚═╝ ╚═════╝
First Threads
Skills, agents, and a sneak peek at Grok Bot
Ship AIPhoenix, ArizonaPress → to begin
Tonight
- 01The brief — thirty minutes of AI news worth arguing about
- 02Skills, agents, and loops — from a markdown file to something that runs itself
- 03Live demos — a thread, a skill, a sneak peek at Grok Bot
- 04What's next — and who made tonight happen
What Ship AI is
Builders showing each other the work.
- Builders & founders
- Phoenix
- Free, always
The best AI education isn't behind a paywall or on a stage — it's in the open, for free.
How the room works
- Craft over hypeThe toolchain, the tradeoffs, the parts that hurt.
- Proof of workA small real result beats a big vague claim. Receipts over adjectives.
- Community-drivenWe listen to the community and build what it says it needs.
Your host
Santos Hernandez
Founder & Host
Founder and Lead Product Engineer building agentic AI systems. Founding product hire at ZBD — the money layer for games — $0 to $12M ARR, the EU's first MiCAR approval, and money transmitter licenses in 26 states and D.C. He started Ship AI to give Phoenix builders a room where you show the work, not talk about it.
Before we start
Join the Ship AI Discord
Everything from tonight lands there — the deck, the skills, the demos, the arguments we don't finish. Free, always.
shipai.club · Free, always
Tonight's room
Workuity
Platinum sponsor · The venue
Every Ship AI session and the October hackathon weekend happen in a Workuity room — this one included. Free for the room means funded by somebody; tonight, that's Workuity.
Community partners
CEI GatewayCommunity partner
Venture Café PhoenixCommunityThe nights and the coffee. Ship AI stays free because these two show up for builders too.
Current sponsors
desicGold · AI deployment
AutomationInterns.comGold · Everyday AISponsors
Ship AI is seeking corporate sponsors
Free for the room means funded by somebody. Published tiers run $1,000 to $10,000 — cash, credits, or hours all count — and we're flexible: there's opportunity at every price point, and a custom arrangement is always on the table.
shipai.club · Free, always
Where we are
- Aug 25 / Aug 26First Threads
- Session 02Daily Driver
- Session 03First Deploy
- Session 04The Night Shift
Four sessions, one business — yours. Each one stands alone, and each leaves a folder the next one reads.
Segment one
The brief
Three stories from the last week — then we write a skill and watch it run.
The brief · 01 / 03
Google launches Gemini Enterprise for finance
- A managed research agent with confidence scores and citations, plus 50+ skills for risk assessment and portfolio monitoring.
- MCP links it to 13 data sources — FactSet, S&P Global, SEC Edgar — inside Workspace and Microsoft 365.
- Early users: CME Group, Deutsche Bank, BNY, Citi Wealth, Lloyds, and Macquarie.
If the skill and the connector are the product, does it matter whose model reads them?
The brief · 02 / 03
OpenAI cuts GPT-5.6 Sol prices 20%+
- Input: $5 to $4 per million tokens. Output: $30 to $20. A promotional rate through Nov 21.
- Covers the pay-as-you-go API, Codex credits, and eligible ChatGPT Work plans.
- Same week Gemini 3.7 Flash got cheaper too, and Anthropic keeps chipping at enterprise accounts on reliability.
A three-month promo, not a permanent cut. Do you build your cost model around this price, or the one it reverts to?
The brief · 03 / 03
Anthropic's run rate hits $65B, IPO looms
- Run rate hit $65B by end of July — Q2 revenue was $11.5B, 14x a year earlier.
- IPO could land as soon as September or October, ahead of OpenAI's — Morgan Stanley, Goldman Sachs, and JPMorgan are running it.
- The target: match or beat SpaceX's record raise, $86.2B including the over-allotment — the largest share sale ever.
SpaceX's raise implied a $1.77T valuation for a hardware company. What does that math imply for one that ships a chat API?
Segment two
Skills, Agents, Loops, and Grok Bot
Before the agent speaks
- InstallThe kit, a language, a project.
- DefineThe agent, written in code.
- ConfigureCloud project, APIs, permissions.
- DeployPush it somewhere it can run.
- TalkThe first useful sentence — after all of that.
Google's Agent Development Kit is a real way to build agents. It is also a lot of machinery before the first useful sentence.
Two ways to get an agent
The ADK way
- A framework to install
- An agent written in code
- A cloud project to configure
- A deploy before it speaks
- The right tool, later
The skill way
- A markdown file
- Instructions in your words
- A folder the agent already reads
- Running tonight, with one word
- The right tool, for this room
Tonight we stay on the file, because that's the one we can finish in this room.
Most people don't need a framework.
They need instructions the model can read.
That's the whole night, said once. Everything after this is that sentence, built live.
What a skill is
- The repeatyou typed it twice
- The filewritten once
- The readevery new thread
- The runone word
The same context paragraph, the same format request, the same correction. If you typed it twice this week, that's the skill.
That explanation, saved as instructions. Not a chat. A file the agent opens before it answers you.
You don't paste it. You don't re-explain. A fresh thread already has it, because the file is in the folder it reads.
You type the name of the skill. It does the job. The test of a skill is whether a blank thread can run it.
A skill is really just a well-formed prompt for one specific problem, saved. We don't invent an example — whatever you re-typed this week is the one we use.
The file
Markdown
A plain text file with light formatting. The format agents read best. A skill is just that file with a bit of frontmatter.
What a markdown file actually is
- Plain textA file you can open in Notes, in VS Code, or in the agent itself. No special app.
- A few marksHeadings, lists, bold. That's most of what a skill ever needs.
- FrontmatterTwo lines at the top, between three dashes: the name, and when to use it. That's how the agent finds it.
- No deploySave it in the skills folder. The next thread reads it. There is no cloud step.
You can open it in Notes. The agent can open it too. That's the whole appeal.
That's the whole file
cat SKILL.md
---name: meeting-notesdescription: Turn a messy call into action items. Use after any meeting.---Read the notes. List decisions, owners, and dates. Stop if there are no notes.
A name, a one-line description, then the steps. Headings and lists if they help. No framework, no deploy, no cloud project.
The description is how it gets found. "Helps with meetings" gets found by nobody. "Turn a messy call into action items" does.
The thread
Start a thread
What separates a thread that produces work from a thread that produces conversation. We start one on screen.
What a good thread actually contains
- Contextwhat already exists
- Rolewho it should be
- Taskone verb
- Formatthe shape you'll use
- Artifactthe file you keep
What you're working on, who it's for, what already exists. Paste the real thing, not a description of it.
Who it should be while it answers. "Someone who has read our last four emails" beats "an expert".
One job, stated as a verb. If it has an "and" in it, that's two threads.
How you want it back — length, structure, the shape you'll actually use.
The file you keep. Everything above this step is setup for this one.
We start one on screen from here. Nobody writes this in order the first time — you write the task, get something generic, and add the other four.
Two ways to use the same box
Conversation
- One question at a time
- Answers you skim
- Starts from zero every time
- Output lives in the scrollback
- You get faster at typing
Work
- A brief, then a draft
- Work you edit and send
- Starts from your own material
- Output lands in a file
- Tomorrow starts where tonight ended
Same box, same model. The difference is whether anything survives the tab closing.
The run
Create a skill
Take the thing you just explained. Write it down as a skill file. Run it fresh with one word.
The thing you explained, written down
/skill-creator
what did you type twice this week?drafting frontmatter + steps…wrote .claude/skills/meeting-notes/SKILL.md
Anthropic's own skill-creator — skills.sh/anthropics/skills/skill-creator. It asks what you keep repeating, takes real examples rather than adjectives, writes the steps as instructions, then we run it in a fresh thread with one word — because a skill that only works in the thread that wrote it isn't a skill.
Follow along or watch; the file is the point, not the typing speed.
Tonight's move
- The threadYou explain the job, once, on screen.
- The fileThat explanation, saved as a skill.
- The blank threadNo context, no help, no preamble.
- One wordThe name of the skill. That's the run.
If it only works in the thread that wrote it, it isn't a skill yet. The blank thread is the test.
The loop
Iterate
A skill is never right the first time. Run it, read what sounded wrong, edit the file, run again.
The skill loop
- 01Run it in a fresh thread
- 02Read what sounded wrong
- 03Edit the file — by hand, or ask an agent to
- 04Run it again
The corrections are the skill. However you make the fix, it has to land in the file — a correction that only lives in the chat is gone by tomorrow's thread.
One word, then a better file
/meeting-notes
no extra promptowners were missing — editing SKILL.mdsecond run: decisions, owners, dates
Same skill, blank thread, no help. Whatever you had to add by hand goes into the file. The third run should not need you.
A skill is never right the first time. That's not a failure — that's the loop.
Loops: a skill on a schedule
- Write itThe same skill file as tonight.
- Trigger itA time, not a prompt — every morning, every Monday.
- It runs itselfNo thread to open, no word to type.
- Result landsIn your inbox, a file, a calendar.
Same file, same skill. The only thing that changed is who presses run.
Loops
A skill runs once.
A goal runs until it's done.
Point it at an outcome instead of a task and it keeps going — checking, adjusting, picking back up tomorrow — until the goal is met, not just the run.
The peek
A sneak peek at Grok Bot
A bot built with the same pieces the room just used — a skill, a markdown file, an agent. Watch it answer live.
The distance is shorter than it looks
- MarkdownInstructions, in your words.
- SkillNamed, saved, read automatically.
- AgentA model that can use the file.
- BotThe same pieces, answering live.
We're going to watch it answer live. If it stumbles, you just learned the loop that fixes it.
Business cases for /goal
- A number, not a stepPoint it at a target — under 200ms, zero flaky tests — and it keeps going until that's true.
- Evidence, not a claimCompletion is checked against real evidence — tests, benchmarks, the artifact — not just a claim.
- Runs when you're not lookingOnly continues when the thread is idle. Picks up tomorrow where it left off.
OpenAI's own examples: latency tuning, flaky tests, migrations, research reproduction. developers.openai.com/cookbook, using goals in Codex.
Thank you
Dan & Workuity
Platinum sponsor · The venue
Workuity Biltmore hosts and sponsors every meetup in this series and the October weekend. Rooms like tonight don't happen without Dan and this community — thank him on your way out.
Thank you
desic
Gold sponsor
AI deployment for businesses — Santos's own team. The same toolchain running tonight's demos is what desic builds and runs for clients.
Thank you
AutomationInterns.com
Gold sponsor
Affordable AI for everyday business — and the newest name on the Ship AI sponsor wall. Welcome Jon S and the interns.
Sponsors
Your logo belongs on this wall
If tonight was useful, this is how it stays free. Every tier is published, every arrangement is negotiable — cash, credits, or hours, at every price point. Talk to Santos before you leave, or scan for the menu.
shipai.club · Free, always
Say hi
- X@5antoshernandezBuild in public — arguments welcome.
- Discorddiscord.gg/kZSJMNveYMWhere tonight's links, slides, and skills land.
- Work with medesic.xyzWant systems like these built and run for you? That's my team.
Everything tonight is open source — the deck, the skills, the template repo. Take it all.
Craft over hype.
Next
Daily Driver — Session 02, Workuity Biltmore
Tonight you wrote a skill and ran it. Next session those skills get a clock — routines, a live calendar, a weekly report that arrives without you asking.
shipai.club · Free, always