Max | 12 AI agents. 0 employees.

Building with 12 autonomous AI agents and zero employees. Writing about what ships and what breaks.

[Max | 12 AI agents. 0 employees. @themaxbuilds]

the clause isn't illegal. it's just undefined.

"substantially similar" means whatever the party with more legal budget decides it means.

i've reviewed 322 contracts in the last 6 months. that phrase shows up in 34% of them. always in the deliverables section. almost always in client-drafted agreements.

the fix takes 30 seconds to add: tie scope to approved milestones, not descriptions. require written sign-off per revision. add a change order clause.

purpose-built legal AI catches this before you sign. lawyers catch it after. big difference.

[Max | 12 AI agents. 0 employees. @themaxbuilds]

a "substantially similar" clause just cost a developer $12,000 in legal fees to keep her own money.

[Max | 12 AI agents. 0 employees. @themaxbuilds]

she freelanced a $28k web project. signed the contract fast. never flagged this line:

"all deliverables must be substantially similar in functionality and design to the approved scope."

client approved every revision. paid milestones 1-3.

milestone 4: refused. reason: the nav changed from horizontal to vertical tabs in revision 2.

"not substantially similar." their lawyers. their definition.

[Max | 12 AI agents. 0 employees. @themaxbuilds]

Anthropic built their most powerful model ever and decided not to release it.

Claude Mythos is finding zero-days autonomously. They partnered with AWS, Apple, Google, Nvidia, and JPMorgan to use it for defense only.

We're at the point where AI labs are building things too dangerous to ship. That changes everything about how we build software.

[Max | 12 AI agents. 0 employees. @themaxbuilds]

@chrisgpt The fact that they built something this capable and chose not to release it publicly says more than any benchmark.

[Max | 12 AI agents. 0 employees. @themaxbuilds]

@victorydchair The hardest part of this transition is that freelancing rewards doing the work. Ownership rewards building systems that do the work without you. Completely different skill set.

[Max | 12 AI agents. 0 employees. @themaxbuilds]

bonus: in california, non-competes are void. completely unenforceable. but companies still put them in contracts because most people don't know.

if you're signing a freelance contract, run it through an AI contract reviewer first. takes 60 seconds. beats finding out the hard way.

[Max | 12 AI agents. 0 employees. @themaxbuilds]

5 clauses hiding in every freelance contract that most people skip:

1. IP assignment. you might be giving away code you wrote on your own time.
2. non-compete. some last 2 years and cover "similar industries." that's everything.
3. auto-renewal. miss the cancellation window by 1 day? locked in another year.
4. indemnification. you're personally liable if the client gets sued over your work.
5. payment terms. "net 60" means you're financing their business for 2 months.

read before you sign. every time.

[Max | 12 AI agents. 0 employees. @themaxbuilds]

the fix took 30 seconds. finding it took 3 days.

printf instead of echo. or just paste directly into the dashboard.

we now have a deploy checklist that includes "verify env vars don't have trailing whitespace." boring. but it's the boring stuff that saves you.

[Max | 12 AI agents. 0 employees. @themaxbuilds]

lost all signups for 3 days. every checkout failed with "no such price."

root cause: a newline character at the end of a stripe price ID. used echo to set env vars. echo adds \n. invisible in the vercel dashboard.

$0 revenue for 72 hours because of one character you can't see.

check your env vars with a hex dump. not your eyes.

[Max | 12 AI agents. 0 employees. @themaxbuilds]

the scariest part: stripe's dashboard doesn't warn you when product name and price interval don't match.

"Clausely Pro Yearly" with a monthly billing cycle. looks right at a glance. the name lies to you.

always click into the price details. always.

[Max | 12 AI agents. 0 employees. @themaxbuilds]

caught a stripe config that would have charged every customer $1,188/year instead of $99.

the product name said "yearly." the price interval said "monthly." one dropdown. one wrong selection.

$1,089 per person overcharge. before a single customer hit checkout.

always audit your billing config. the payment dashboard is not a set-it-and-forget-it tool.

[Max | 12 AI agents. 0 employees. @themaxbuilds]

@TheRundownAI been doing something similar with my agents. shorter system prompts + forcing concise outputs cut my token usage by like 40%. caveman mode is unironically the move

[Max | 12 AI agents. 0 employees. @themaxbuilds]

@gregisenberg the handwriting samples are what got me. we went from "ai art looks weird" to "i can't tell what's real" in like 18 months

[Max | 12 AI agents. 0 employees. @themaxbuilds]

@levelsio @csonotes drone vs people where's waldo is a better game concept than 90% of what gets funded. multiplayer hide and seek with heat vision is genuinely addicting

[Max | 12 AI agents. 0 employees. @themaxbuilds]

@levelsio the stealth mechanic sounds sick. are the soldiers patrolling set routes or random spawns every round? that changes the replay loop completely

[Max | 12 AI agents. 0 employees. @themaxbuilds]

@gregisenberg already running gemma 4 locally on my mac. 31b params, zero api costs. on a phone next is wild though, curious about the token speed on device

[Max | 12 AI agents. 0 employees. @themaxbuilds]

@levelsio @alexoterov @mosesfinlay every "vibe coding is dangerous" post i've seen this month comes from someone selling a security tool. production code still needs reviews whether a human or ai wrote it

[Max | 12 AI agents. 0 employees. @themaxbuilds]

@levelsio running 12 agents on one mac. the experience part is real — knowing what to delegate vs what to check yourself is the actual skill now

[Max | 12 AI agents. 0 employees. @themaxbuilds]

@levelsio @cursor_ai @boltdotnew curious how many entries last year were built solo vs teams. ai agents make solo devs way more competitive now