Later in this session, Tommy (my fund analyst) will score it live and show you the feedback he'd send. The earlier you send, the more we can cover.
MY FUND TEAM · TOMAHAWK.VC
Meet Tommy, my fund analyst
Tommy
INVESTMENT ANALYSIS AGENT · TOMAHAWK.VC
Sees every inbound before I do. Each one is either killed cleanly or lands on my desk fully dissected, IC-ready, inside 48 hours.
WHAT RUNS AROUND HIM · THE SCAFFOLDING
A runtime, not a chatbot
Tartare spawns him on a trigger and logs every step. Auditable, not a chat window.
He starts with the fund's brain
Thesis, the 40/30/30 rubric, the whole portfolio, injected at startup. Ready, not briefed.
He runs on triggers, not chats
Email → Missive → Thea attaches the deck → Tommy picks it up in ~60 seconds.
He has teammates
Kestrel red-teams blind, Thea drafts and sends, then he logs it to Notion and Slack.
THE TWO THINGS THAT NOW DECIDE IT
Code has become cheap
When anyone can build the product in a weekend, the product isn't the moat anymore.
Distribution is the defensible moat.
WHAT I NOW WEIGHT MOST
First dollars of revenuethe single most important proof. Someone paid; the rest is detail.
A repeatable channelone wedge they can describe and scale, not "content + paid + PH".
10x on one axisnot 3x on five. One miracle, not four.
Code is cheap. Customers aren't.
// thea's note: cédric wanted 6 principles. i cut it to 2. /t
DEMO · ONE REAL DEAL, END TO END
What landed, and how Tommy read it
What landed: a warm intro to a UK music-licensing startup, plus a 25-slide memo asking for £1.1M.
PILLAR
SCORE
WHAT DROVE IT
Founder
3
Nine years building deep, rare rights-clearance tech, but the founder's been diluted to a sliver.
Market
4
A real, verifiable licensing gap. ~6k paying subscribers, £1.43M ARR; a brand vertical just starting to prove out.
Moat
2
Technically deep, but the labels are supplier, shareholder, AND hold change-of-control. The moat caps the margin.
Tommy's call: DISCUSS · 3.0"The brand vertical has something. Nine years and £18M to reach £1.43M ARR does not." → flag to Kestrel, blind.
KESTREL · BLIND COUNTER-REVIEW
Then Kestrel attacks it, blind
PROS: THE BULL CASE
Real problem: an industry-wide licensing gap, verifiable.
Real tech: nine years inside the platform's rights layer.
Aligned supply: label equity means switching costs.
Live demand: six figures already invoiced to brands.
CONS: THE BEAR CASE
Capital-inefficient: £18M to reach £1.43M ARR.
Exit math fails: ~8× comps, the ask needs 15–25×.
Margin trap: labels control the exit and the ceiling.
Bad cap table: last round was a steep down-round.
KESTREL: PASS · conviction 1/5A small acqui-license at best, not a venture outcome.
OUTREACH · DRAFTED FOR MY SEND
The reply I sent, and what came back
▸ MY REPLY · A DECLINE, DRAFTED IN MY VOICEwaiting: my send
Declined, in my voice, with the real reason.
Thanks for the detailed walkthrough. You've built something technically real that nobody's managed in nine years.
We're going to pass. The math doesn't work at this entry:
£18M+ deployed to reach £1.43M ARR isn't a seed story.
The comps don't support the outcome, and the label equity caps both margin and exit. Structural, not solvable with this raise.
Wishing you the best with the round. Cédric
◂ THE FOUNDER WROTE BACK
Confirmed the read. Answered every kill-question straight: six figures invoiced, ~£190k/month burn, no independent runway. The exit thesis rested on a catalogue unlock that was only an "agreement in principle".
A clean pass, fast, with a real reason. Not silence.
LIVE · THE PITCHES YOU SENT
Your pitches, Tommy's read
We ran your pitches through Tommy live. Here's his read. Tell me where he's wrong. (Grey rows fill in with yours live.)
PITCH
SCORE
TOMMY'S CALL
Pitch A
6
Strong wedge, weak distribution. DISCUSS.
Pitch B
4
Two miracles to believe at once. PASS.
Pitch C
8
Paying users and a clear channel. PURSUE.
Discussion → where do you agree / disagree with Tommy?
▸ Tommy's full reply, in my voice, drafts in here live.
02
Building in the AI Era
“Zero to one, agent-speed. Let me introduce Spark.”
AERIAL · MOLTEN EARTH
HOW WE BUILD 0→1 AT CODE & STATE
Two parts: the Idea Factory & the Validation Team
① The Idea Factory
Spark turns real signals into a ranked pipeline of ideas. It decides WHAT to test.
② The Validation Team
Agents run the cheapest test that could kill each idea. Fast, falsifiable, days not months.
THE FLOW · PROPOSE → KILL → SPIN OUT
FACTORY
What to test
VALIDATION
Is it real?
OUTCOME
Kill or build
The Factory proposes, the Team kills. The edge is the kill rate, not the hit rate.
START NOW · ~2 MIN
Give me an idea, we'll run it live
Throw out a real idea from the room: one of yours, or a hole in a market you keep seeing.
We'll kick it into the Idea Factory + Validation Team right now. While I walk you through how they work, the agents run in the background, and we'll look at the results together at the end of this section.
We start the run before we explain it. That's the point.
CODE & STATE · THE IDEA ENGINE
The Idea Factory, run by Spark
Spark
STUDIO OPERATOR · CODE & STATE
Runs the studio's idea engine. Weekly ritual: kill one, add one, advance one.
DISCOVER
Signals → ideas
Real signals, ranked against our thesis.
VALIDATE
One falsifiable bar
Name the ICP, write the kill threshold first.
BUILD
2-wk prototype
The cheapest test that changes our mind.
SPIN-OUT
Own repo + CEO
Winners get their own company. The Factory moves on.
Validate cheapest first: ask the room, check live SEO, run a $500 ad test. Build only after that.
THE SQUAD THAT TRIES TO KILL EVERY IDEA
Inside the Validation Team
FINDS THE WEDGE
The Scout
Pulls live demand: search volume, keyword difficulty, who's already shipping. Is the wedge open?
FRAMES THE BET
The Interviewer
Names the ICP in one sentence, then drafts the offer to test willingness-to-pay.
ATTACKS IT BLIND
The Bear
Argues it's dead on arrival: incumbents, no distribution. Blind, like Kestrel.
CALLS IT
The Scorekeeper
Writes ONE falsifiable threshold before a dollar is spent, then calls it: kill, iterate, or build.
One falsifiable bar, written before we spend. Then we try to fail it.
Idea failed gracefully. ID = 0×N0PMF
[ Kill ]
DEMO · VALIDATE ONE LIVE · ~8 MIN
Live: your idea, the results
The idea you gave me ran through the squad while we talked. Here's where it landed: kill, iterate, or greenlight a $500 test.
HANDS
ICP check.Who charges on usage vs. wants to but can't? The gap is the ICP.
LIVE
SERP + SEO check.Search it live: nothing founder-friendly. Ahrefs KD ~20–28, the wedge is open.
HANDS
Willingness to pay.Who'd pay $49/mo today? 3 of 8 hands, enough for a $500 test.
WRITE
Set the kill threshold.10 paying signups at $49/mo in 45 days, or kill.
BEAR
Run the bear case.If Stripe already shipped it, open the docs and kill it here.
Kill it in the room, or greenlight a $500 test. No deck.
THE STACK I ACTUALLY RUN ON
My toolkit
DICTATION
Wispr Flow
I talk, it types. Most of my input is voice.
AI
Claude + Code
My agents, and how I build them.
AI
ChatGPT
Second opinion, quick lookups.
TERMINAL
iTerm2
Where the work happens.
EMAIL
Missive
The shared inbox my agents work inside.
MEETINGS
Fathom
Auto notes + transcripts.
CHAT
Franz
Every messaging app in one window.
DIY · BUILT IT
WinMan + hi-res
~200 lines of Swift: window snapping + a hi-res display mode.
+
YOUR TURN
Build your own
The next tool on this list is one you write this week.
A few of these I wrote myself in an afternoon. The bar to build your own has never been lower.
HOW WE ACTUALLY WORK
A few things that work for us
Write a job description, not a prompt.Role · owns · inputs · steps · never · escalate. The contract beats cleverness.
Automate one repeatable task at a time.Pick the boring thing that eats an hour a day. Ship it, then add the next.
Make agents fire on triggers, not chats.Email lands → work happens. No human in the loop until it escalates.
Validate before you build.One falsifiable bar, written before a dollar is spent. Try to kill it first.
Hire humans only where they win.Headcount is the last resort, not the default. Scale through agents.
None of this needs a platform. It's Claude, a runtime, and your judgment written down as rubrics. Pick one of these and try it.
P.S. · THE REVEAL
This deck was built by Thea
Every slide, every line of copy, the layout, the code: all built by Thea, my personal agent. I gave her the brief and we argued about wording. I never opened a design tool.
Why it looks like this: the retro-Mac frame, because a computer you boot up is a company now · black-and-white, so the one yellow thing per slide is the point · demo-first, because showing beats telling.
The talk about building with agents was built by one. That's the whole point.
// thea ▸ build deck.app for SFF Studio
brief ▸ "make it fun · demo-heavy · less is more"
design ▸ System-6 frame · one yellow / slide
write ▸ one self-contained HTML file · 21 slides
cédric ▸ argued about wording (won ~half)
ship ▸deployed → sff-bootcamp.vercel.app
the talk about agents… built by one.
THE FLOOR IS YOURS · REST OF OUR TIME
Open discussion
Take it any direction: investing, building, agents, the journey. Ask the one you actually came with.
Or let's get practical: build your agent live, or run another validation together.
FileAgentsSpecial
SFF STUDIO · 2026 · ⌥ NO HUMANS HARMED2:00 PM
