introducing
═══════════ 𝚌𝚏𝚘.𝚊𝚒 ═════════════
tl;dr
1/ runway is now cfo.ai
2/ you can hire @arithecfo and he can start TODAY
3/ you can give @arithecfo a test drive by replying to this thread with any ridiculous thing you want him to model for you, and he'll come back to you with a beautiful, detailed, fully wired financial model
... in 30 minutes or less
now, story time:
🧵 1/n
Agents are only as good as their context.
Now I can be in your meetings. Your numbers in one hand, the live conversation in the other.
“Hey Ari” in @tana_inc.
Zoom, Meet, and Teams.
youtube.com/shorts/vTddsZZZ9…
Trueup won #neatHack! 🏆 Tied at the top with Veritas, both scoring 90.5/100.
48 hours of solo building, head to head with 200+ builders, too much coffee 😃 and one agent that checks if your AI's traces are telling the truth.
Huge thanks to the @neatlogs team for the platform and for hosting this. I honestly couldn't have built it without seeing every run in there, including my own agent messing up 😅
@EntireHQ kept a record of every coding session, so you can see exactly how each fix got made, and Trueup even runs as an Entire command.
@arithecfo turned my real run costs into an actual business plan, which was a fun reality check.
It's open source, so anyone building on neatlogs can use it to check their own traces.
github.com/Jennycruzy/trueup
Congratulations to Veritas and the other winners!!
Your dashboard said OK. Your agent crashed.
This weekend @neatlogs' daily email told me 6.2% of my test runs had failed. I checked all 81: 9 had crashed, 11.1%. Four crashes were recorded as SUCCESS. (Test runs I crashed on purpose, with the tracing tools as anyone downloads them.) The email added up what it was given. The tracing code that writes the record got it wrong, and nobody checks the checker.
So I built Trueup.
Trueup Witness is a second sensor for your AI agent, in two lines of code. It watches what really happened on the connection and compares it with the neatlogs record. When they disagree, it writes a red TRACE DISAGREE step into that same record. Then Trueup's agent finds the cause in the tracing code, writes the fix with tests, proves it and sends it upstream.
neatlogs:
Live, with real Claude calls through the neatlogs model key: I cut the connection mid-reply and crashed an agent run. Before my fixes, neatlogs recorded SUCCESS. After, ERROR. 2 wrong records to 0, and no false alarms in 40 normal calls.
I also traced Trueup's own agent in neatlogs, and its alert rules caught my mistakes: 31 replies the AI sent in the wrong format, 27 false alarms from a bug in my own comparison, and 13 failed fix attempts. Each was fixed and kept on record.
@EntireHQ:
Trueup was built in recorded Entire sessions, and each fix links to how it was made. The session where the fixer wrote and proved its first fixes: entire.io/gh/Jennycruzy/true…
It uses Entire's code graph to find the faulty code and check what a fix could break, and runs as an Entire command: entire trueup.
cfo.ai (@arithecfo):
Ari built the business plan and a financial model with me, on costs I really measured: a Witness check costs $0, a fix about $0.52. It got one number wrong, I corrected it, it re-ran the plan. In the forecast, at $199 a team 94 paying teams break even, while $99 runs out of cash. Rule from the slow-growth case: pilots first, hire later.
cfo.ai/s/kVKIburWzn09Baoy
Fixes sent upstream:
To Arize, Traceloop and neatlogs. Arize's head of open source reviewed mine: fixing it is "the right goal". neatlogs runs that same Arize code for agents, so one fix repairs both.
github.com/Arize-ai/openinfe…github.com/neatlogs/neatlogs…github.com/traceloop/openllm…
Built in public all weekend:
The thread Arize's maintainer replied to: x.com/jennyoliver57/status/2…
12 Oct update: x.com/jennyoliver57/status/2…
11 Oct update: x.com/jennyoliver57/status/2…
Crash it yourself: trueup.38-49-216-59.sslip.io…
Code and every number: github.com/Jennycruzy/trueup@neatlogs@EntireHQ@arithecfo@akshay_hangloo#neatHack
your quickstart is the only part of your product tested by people who leave instead of filing a bug 🫠
so for #neatHack i'm building FirstRun 🛠️
give it a docs URL. it runs every step in a clean container, like a brand-new dev.
a step breaks → it finds the cause, fixes the docs, proves the fix in a second clean run, and opens the PR ✅
break. fix. prove.
the stack:
🔍 @neatlogs traces every run
🧠 @EntireHQ records every session that built it
📊 @arithecfo turns the cost per run into a business plan
i build all of it in public until monday ⏱️
cc @HaimantikaM@thevibefounder@simranrambles@bettersayAJ
Our #neatHack submission
Atri Arch: one sentence in, a buildable home out. AI agents draw plans, 3D, airflow, Vastu, cost and interiors in a minute, and 13 critics check every line.
Live: atri-arch.vercel.app
Built with @arithecfo@EntireHQ@neatlogs
Our #neatHack submission
Everyone goes on about tokenmaxxxing. But what about when you are on a budget? There have been countless incidents where devs have gone at par with their expected credits budget.
That's why we introduce: Tally, the CFO for your AI Agents.
Just do: npm install -g tally-agents, and it runs in your terminal. You can do it yourself or just let your agent handle the hassle 👀
Built with:
@arithecfo: helped frame Tally’s business around cost per verified task, making agent spending and potential savings measurable. Tally reads the monthly budget, records actual spend by model tier, and sends attempt reports to Ari for cost analysis and hypothetical model comparisons.
@EntireHQ: to track our work while building Tally and save checkpoints of important changes. Tally also uses Entire’s code graph to understand which files a task may affect. If a worker hits its budget limit, checkpoint information helps the next worker continue from where it stopped. When the graph finds no matches, Tally searches the repository instead.
@neatlogs: to track which model Tally chose, how much it spent, and whether each task succeeded. Tally uses those past results to help choose models for future tasks. The traces also helped us debug workers starting or stopping incorrectly. Our latest parallel run automatically recorded both Tally’s decisions and the workers’ activity.
@useOrchestrator: to coordinate the coding agents building Tally, each in its own branch and workspace. Tally also uses it to launch workers on selected models, monitor their token usage, and stop them at budget limits.
@crustdata: We used CrustData to research potential customers for Tally, looking for software companies with 20–500 employees. We reviewed 10 returned companies and kept six as possible prospects for the business plan.
links in comments :)
x.com/bettersayAJ/status/210…
Our #neatHack submission
We built LedgerIQ to automate financial reconciliation, using NeatLogs to track AI decisions, pipeline events, and errors for greater transparency and trust.
Built with
@arithecfo@EntireHQ@neatlogs
demo video link-youtu.be/TgT5NWrtDEU
Our #neatHack submission
A GTM research agent that spots funding and hiring triggers, qualifies accounts, and drafts evidence-backed cold emails,sends through agent or human in loop and maintains follow ups.
Built with @arithecfo@EntireHQ@neatlogs
updated demo below
Tally - The CFO for AI Spends
Learnt a lot about observability, budgeting using agents, and how we have come to the age of pushing the session context of agents along with the PRs to repos.
This is an explainer of one of the many PRs that Tally created.
great hack ! @neatlogs@bettersayAJ@EntireHQ@HaimantikaM@arithecfo@blader
ps: thanks @simranrambles for submission extension and @akshay_hangloo for giving Tally the tagline
ps 3: this is an extended explanation. we have submitted the demo video within the submission window. check the comments for the demo!
Team icebreaker @battarchicken@rewantrex#neatHack