Talk by Ramana Siddanth Emani (Auditoria AI) arguing that the bottleneck in shipping production finance agents is the developer in the loop, not the model or hardware. He describes reusing agent harnesses to automate the developer loop itself: coding agents running in parallel across separate git worktrees, with skills enforcing the right patterns. Tasks are sourced from QA reports, Jira tickets, and GitHub pull requests; sub agents pull traces and logs, write and run end-to-end tests, build, and report back, involving the developer only at key steps. He covers stepping out of the loop safely with a human as final verifier, and how this applies at Auditoria in finance, where agents talk to agents and reconcile source data. # Video Identity Observations — "Your Finance Agent's Bottleneck Is You — Ramana Siddanth Emani, Auditoria AI" ## Explicit Self-Description by the Creator - **Speaker affiliation**: > "my name is Siddhant Imani and I'm a data scientist at Auditoria AI" - **Company scope claim**: > "we build production agents for finance" - **Venue context (stated)**: > "this AI engineer's world fair" — a conference talk situated between tracks: > "This talk is in between the harness engineering track and AI for finance" - **Corrected speaker name**: The video title says "Ramana Siddanth Emani," while the transcript self-introduction is "Siddhant Imani" — a discrepancy between title and transcript. ## Intended Audience (as addressed in the talk) - **Developers building agent harnesses**: > "I'm here to talk about the harnesses that you guys are developing and using these internal harnesses to build your production agents" - **Developers seeking productivity gains**: > "if you're a developer yourself, how do you be 10x productive with the agent harnesses that you're using" - **Finance executives (secondary, business-development overture)**: > "if you're a CFO in the audience, I would love to speak to you after the session" - Interactive audience polling is present: > "how many of you know you can give goals to your agents?" ## Claimed Scope of the Channel's Subject Matter (as demonstrated) - **AI agents in production finance software**: bugs, compliance, QA pipelines — e.g., discussion of SOX compliance and audit review: > "you have a human auditor which reviews the code and you have a controller which signs off under your socks compliance" - **Agent harness engineering**: sub-agents, parallel work trees, skills, MCP tools, orchestration, and "minimal UX." - **Developer autonomy via recursive self-improvement**: goals, loops, and "dreaming" agents. ## Observable Presentation Patterns (not claims) - **Conference-talk format**: greeting, session framing, Q&A with audience, closing thanks and applause indicate a live presentation rather than a studio production. - **Provocative but self-aware framing**: > "sorry for the rude title. I don't mean to call the audience here the bottlenecks" - **B2B/enterprise posture**: the CFO comment and finance-regulation discussion demonstrate a dual audience appeal — technical practitioners plus potential enterprise customers. - **Prescriptive, list-driven teaching style**: "four primitives" (sub-agents, skills, MCP tools, minimal UX) delivered as actionable guidance: > "at the end of the session I hope you can 10x your production code" ## Evidence Gaps / Limits - No channel name, upload cadence, subscriber counts, or curated metadata were supplied — audience and scope observations rest solely on this single talk's transcript and title. - The talk mentions model names ("Fable 5," "Mythos 5," "GPT 5.6") without verifiable context; these are treated as speaker statements only. # Delivery Observations — "Your Finance Agent's Bottleneck Is You" ## Tone - **Conversational and casual**, with frequent filler words: "you know" ([76], [315]), "um" ([370], [681]), "let's say" (repeated at [269], [291], [568], [591], [601], [611]). - **Self-aware and disarming opener**: > "So, sorry for the rude title. I don't mean to call the audience here the bottlenecks" ([12–23]). - **Light humor** toward the end: > "And you can just book a vacation maybe or work from home" ([623–636]). ## Vocabulary - Heavy use of **developer/agent jargon**: "harnesses," "work trees," "sub agents," "TDD," "PR," "MCP tools," "SOX compliance." - Recurring numeric specificity: "50 active work trees," "48 GB of RAM" ([269]), "five to six bug tickets" ([601]), "3 and 1/2 months" ([130]). - References to **hypothetical model names** — "Fable 5," "Mythos 5," "GPT 5.6" ([120], [556–568]). ## Recurring Phrases - **"Work trees"** repeated many times ([201], [212], [269], [291], [303], [327], [746]). - **"Minimal UX is the key"** / "minimal UX" appears at least three times ([269], [418], [766]). - **"So,"** as a sentence-starter is pervasive throughout ([87], [98], [109], [153], [212], [476], etc.). - **"Let's look at"** as a transition device ([269], [315], [476], [546]). - **"Let's say"** used repeatedly to introduce hypotheticals. - **"At the end of the day"** ([257]) and **"at the end of the session"** ([187], [746]). ## Teaching Habits - **Rhetorical questions to structure content**: "So, what is the problem?" ([109]), "So, what is the answer?" ([130]), "So, where do these tasks come from?" ([279]), "What is the bottleneck?" ([523]). - **Audience interaction**: direct question — "how many of you know you can give goals to your agents? … Anybody?" ([647–657]), and "I would like to ask a question in the audience" ([370]). - **Enumerated structure**: "I'm talking about four primitives here" ([175]), then explicit parts — "Let's look at part two" ([546]). - **Concrete walkthrough example** of a 9-step QA-to-deployment pipeline ([315–370]). - **Recap/summary at the close**: "So, just an overview of what I've covered till now in this session" ([746]). - **Named framing devices** for concepts: "skills… your organization secret recipes" ([212]), "work trees as isolated folders" ([201]), "dreaming" for background analysis ([693]). - **Sales-adjacent aside**: "if you're a CFO in the audience, I would love to speak to you after the session" ([44]). ## Communication Patterns - **Parallel triads**: "Is it the model? … Or do you need faster GPUs? Or do you need a better framework?" ([109–130]). - **Repetition for emphasis**: "writing code is very easy" ([87]) and "writing code is easy" ([142]). - **Physical/relatable analogies**: "if you squint your eyes and see" ([418]), "the number of neck rotations to ship one change" ([452–462]), comparing agents to human dreaming ([693–714]). - **Restatement of core thesis** as a memorable principle: > "Always have the human as a verifier, but not the throughput ceiling because human attention is very limited" ([777–789]). - **Signposting the track positioning** early: "This talk is in between the harness engineering track and AI for finance" ([44–55]). ## Stable Patterns vs. Isolated Moments - **Stable**: "So," as connector, "let's say" hypotheticals, rhetorical question transitions, minimal-UX messaging, jargon density. - **Isolated**: the "[snorts]" moment mid-sentence ([714]), the trailing, slightly fumbled close — "Thank you for your Thank you for I hope you um learned something" ([789–801]) — and the "book a vacation" joke. Insufficient repetition to treat these as habits. ## Uncertainty Notes - Some transcript gaps/speaker-attribution oddities (e.g., "a You can have an army of them" at [201]) limit confidence in exact phrasing. - Model names ("Fable 5," "Mythos 5," "GPT 5.6") may be transcript artifacts or intentional hypotheticals; intent cannot be confirmed from delivery alone. # Observation Document: "Your Finance Agent's Bottleneck Is You" ## Session Overview - **Speaker:** Ramana Siddanth Emani (introduced as "Siddhant Imani"), data scientist at **Auditoria AI**, which builds production agents for finance. - **Framing:** The "bottleneck" refers to the developer/harness, not the audience personally. Talk sits between a harness engineering track and AI for finance. - **Goal stated:** Help developers be "10x productive" using agent harnesses. ## Core Claim: Demos Fail in Production - Beautiful demos are easy to ship, but **production bugs are high and production guards are built "by the hour"** — stated as "a hard fact." - Agents have never seen new customer data when promoted from demo to pilot. - Speaker dismisses model/GPU/framework swaps as real fixes: new models every ~3.5 months, new chips yearly, new frameworks daily — so the answer is **dev loop velocity**. ## Four Primitives 1. **Sub-agents** — spawn "an army"; **git work trees** as parallel isolated folders so agents don't fight over the same work. 2. **Skills** — organizational "secret recipes" that ensure agents use correct workflows. 3. **MCP tools** — agents connect to any third-party system; client data can live anywhere. 4. **Minimal UX / orchestration** — many sub-agents create orchestration work; compress into one pane of glass. ## Sub-Agent Orchestration Details - Claim: 48 GB RAM MacBook supports ~**50 active work trees / 50 sub-agents** in parallel. - Jira tickets mapped to separate work trees, each handled by an agent that can spawn sub-agents. - "Don't queue up tasks — agents do that better than you." ## Example Pipeline (Steps 1–9) Bug ticket → parse requirements → root cause analysis → pull traces/logs → work tree → TDD → implement fix → local E2E testing → PR → review → merge → Docker build → dev deploy/test → stage deploy/test → return to QA. - **Human contact points claimed:** only steps 1 (observe agent work) and 9 (validate after staging). "The in-between steps, the agent can do a lot better work." - **Minimal UX widget:** single macOS widget showing Kubernetes services, pods, logs, Jira tickets, GitHub PRs, and a cloud code session — reduces "neck rotations" across multiple monitors. ## Finance-Sector Qualifications - Finance has heavy regulation; without AI: human auditor reviews code, controller signs off under **SOX compliance**. - **Accountability tension (unresolved):** "Where do you keep the accountability? If something goes wrong in production, you can't say 'Claude is doing this, something is wrong.'" - "Moving fast and breaking things" in finance is very different; bottleneck becomes **human attention**. ## Part Two: Removing Yourself from the Loop - With recursive self-improvement (speaker references "Fable 5 and Mythos 5 and GPT 5.6"), production failures become input. - **Self-improvement loop:** run 1–2 days solving 5–6 bug tickets → agent analyzes bottlenecks → removes them daily → after a month, "fix this bug for me" in one sentence, agent handles end-to-end. Speaker jokes about booking a vacation. - **Goals + loops:** set a goal (e.g., data discrepancy in a report), close the laptop, monitor from phone. - **"Dreaming":** agent compacts recurring customer session patterns into data points to upgrade itself. ## Conclusions - Sub-agents + parallel work trees + skills + MCP tools + one pane of glass + goals/loops for autonomy. - **Key closing principle:** "Always have the human as a verifier, but not the throughput ceiling because human attention is very limited." - Working 9–5 writing code "is not valid anymore." ## Tensions Preserved - **Human role contradiction:** speaker says human required at steps 1 and 9, then advocates "removing yourself from the loop" — resolution offered is human-as-verifier-only. - Accountability in regulated finance is raised as a question but not resolved. - Many claims (50 work trees, one-month self-automation) are stated without evidence or qualifications in the transcript. # Speaker Attribution Record — Video `01a15ca8-2f54-78d5-b909-82a547b7ac18` ## Video **Title:** *Your Finance Agent's Bottleneck Is You — Ramana Siddanth Emani, Auditoria AI* ## Speakers ### Primary Speaker — **Siddhant Emani** (self-identified as "Siddhant Imani") - **Role:** Creator / presenter (talk speaker) - **Affiliation (stated by speaker):** Data scientist at **Auditoria AI**, which "builds production agents for finance" - **Talk framing (per speaker):** Sits between the "harness engineering track" and "AI for finance" ### Other Voices - **Audience members** — brief interaction; the speaker asks "how many of you know you can give goals to your agents?" and notes a response ("Nice"). No individual audience speaker identified. - **Unidentified speaker(s)** — transcript ends with "Hey." after applause; identity unclear from source. ## Promotional Segments - Speaker invites **CFOs in the audience** to speak with him after the session ("if you're a CFO in the audience, I would love to speak to you after the session") — flagged as promotional/networking outreach for Auditoria AI. - Repeated references to Auditoria AI's domain (production agents for finance, SOX compliance context). ## Provenance Facts - **Third-party products/models referenced (no endorsement claims recorded):** "Fable 5," "Mythos 5," "GPT 5.6," GPUs/chips, MCP tools, Jira, GitHub PRs, Docker, Kubernetes, macOS widgets, cloud code sessions. - **Claimed techniques described by speaker:** sub-agents, work trees, skills, minimal UX, goals/loops, "dreaming" (session compaction). - **Quantitative claim (speaker's own, unverified):** "with 48 GB of RAM on your MacBook, you can have 50 active work trees." - **No external cited sources** are named in the transcript. - **Comment availability:** not stated in supplied material. - **Curated metadata:** none supplied beyond title; title names "Ramana Siddanth Emani" while speaker self-identifies as "Siddhant Imani" — discrepancy noted, not resolved. ## Uncertainty - Exact spelling of speaker's name differs between title and self-introduction. - Final "Hey." speaker identity unknown. \>\> Hello everyone. Welcome to this session about your finance agent's bottleneck is you. So, sorry for the rude title. I don't mean to call the audience here the bottlenecks, but I'm here to talk about the harnesses that you guys are developing and using these internal harnesses to build your production agents. So, my name is Siddhant Imani and I'm a data scientist at Auditoria AI and we build production agents for finance. So, if you're a CFO in the audience, I would love to speak to you after the session. This talk is in between the harness engineering track and AI for finance. So, this talk is mostly about identifying the bottlenecks within your developer and if you're a developer yourself, how do you be 10x productive with the agent harnesses that you're using. So, all of us have seen, you know, beautiful demos in this AI engineer's world fair. But, once these demos are promoted to pilots and you start onboarding new customers, the agent has never seen these future data. So, all of us know production bugs are very high and production guards built by the hour. So, that's a hard fact. And writing code is very easy. So, shipping beautiful demos and showing it to a lot of people is very easy right nowadays. So, what is the problem? And why do these demos fail in production? Is it the model? Do you need a better model? Fable 5, perhaps? Or do you need faster GPUs? Or do you need a better framework? Maybe. Or your RALF loops are not working properly. So, what is the answer? If you wait 3 and 1/2 months, we are awarded with a new model in the market. So, we can easily swap models. If you wait perhaps 1 year, we have new chips. We have faster GPUs. And again, writing code is easy. So, we have new from frameworks every day. So, you can swap your framework every now and then. So, how do we in real time fix these production bugs? The answer is your dev loop velocity. The model capability increases very exponentially. And the developers have to spend a lot of time every day to automate your developer loop. So, I'm talking about four primitives here. All of you need to think about loops. And at the end of the session, I hope you can 10x your production code. So, first we have sub agents. Nowadays, whatever harness you're using, you can spawn new sub agents. You can have a You can have an army of them. And get work trees are your best friend. So, think of work trees as isolated folders. And inside these folders, the agent writes whatever code it's generating. So, you want these work trees to be in parallel. So, the sub agents are doing independent tasks and are not fighting over the same thing. Second, we have skills. These are your organization secret recipes. So, make sure you have a lot of skills because these skills, once you start say giving it to your agents, the agents will always make sure to use the correct and proper workflows to solve whatever production bug you're facing. And of course, all of us have seen a lot of MCP tools being shipped into the market right now. Everybody says we can, yeah, the agent can connect to whatever MCP tool and whatever third-party server there is. And your client data can live in any system you want. And at the end of the day, if you have a lot of sub-agents, you have a lot of work to orchestrate. So, minimal UX is the key here. Let's look at the sub-agents. With you as the orchestrator, you can have, let's say, with 48 GB of RAM on your MacBook, you can have 50 active work trees. That is 50 active sub-agents working independently on different tasks. So, where do these tasks come from? So, let's say the production software you're going to ship has a lot of bugs that your QA is reporting. So, all the Jira tickets can be thought of in a separate different work tree. So, different work trees are handled by a separate agent, and these agents can spawn multiple sub-agents to solve that particular task. You don't want to queue up your tasks because the agent is will do that a lot better than you. Let's look at, you know, um, an example harness. What if the QA reports a lot of bug tickets, and somehow magically there is an agent which parses the requirements, does a root cause analysis, pulls all the traces, pulls all the logs, puts all this in a separate work tree, does the TDD, does the, implements the fix. Because it's in your local system, you have to do test scripts, local end-to-end testing. You create a PR. You submit the PR to your team for review. And after review, you merge it into your master branch, let's say. After merging it, obviously, you have to build a Docker image, deploy it into your development environment, test it, again ship build an image to your stage environment, test it, deploy it to stage. And then you go back to the QA saying, "Here you go. You can test it now." So, I would like to ask a question in the audience, um at what points do you think the human contact is required in this, um steps 1 to 9? So, I would say the human is only required at steps 1 and 9 because the in-between steps, the agent can do a lot better work. There needs to be a human to see what work the agent is doing. And then needs to be a human at the end to validate after the work is being shipped to stage. And obviously, we need minimal UX because humans love minimal UX. So, in the image, if um you can if you squint your eyes and see, the image shows um the production agent software that you're building, the project dashboards which shows all your Kubernetes services, pods, examples, all the logs, system logs, all your Jira tickets, all your GitHub PRs, and maybe a cloud code session at the bottom. So, this is basically a macOS widget, and you don't need to open multiple windows to do all of this work. A developer does like variety of things in their software development life cycle. So, you can use just this one widget to do a lot of things. So, you can see from the graph also, the number of neck rotations to ship one change like reduces a lot drastically. And I imagine all of you have like two to three monitors on your table and you just keep rotating your neck orchestrating these agents. So, Auditoria works in finance. So, there's a lot of regulation and policies happening in finance right now. So, what does it look like for orchestrating a team of sub agents in the finance sector? If we take AI out of the picture, usually what happens is you have a human auditor which reviews the code and you have a controller which signs off under your socks compliance. And reviewing agent to agent, it it doesn't Where do you keep the accountability? If something goes wrong in production, you can't say Cloud is doing this. Something is wrong. So, but let's say you have all these sub agents and you're using this harnesses to fix bugs in real time. What is the bottleneck? It becomes a human attention because you yourself have to orchestrate all these different tasks. And moving fast and breaking things in sector in the finance sector is a lot different. So, let's look at part two, which is removing yourself from the loop. Till now I've been saying a human is required to see what the agent is doing and at the end also to validate what the agent has done. But with the self-improvement of the agent and model capabilities these days, we get Fable 5 and Mythos 5 and GPT 5.6 also. So, what does it look like when you have this recursive self-improvement in your internal developer developer harnesses? So, all your production failures become input. So, let's say you automate keep automating these cell developer harnesses every day and you ask the agent to upgrade itself essentially. So, you do a task. You let the loop run. Let's say one or two days. You solve five to six bug tickets. And you just tell the agent to analyze all the bottlenecks in this process. Make a list of them. And somehow slowly keep removing these bottlenecks every day. At the end of one month, let's say, you have a really nice self-automated loop where you just type in one sentence and just say fix this bug for me. And the agent goes off, connects to all your database systems, fetches all the logs, traces, tickets, and ships it and migrates it migrates it to the Jira to QA pipeline. And you can just book a vacation maybe or work from home. And what does it look like internally and what happens when you stare less and ship more? Nowadays, how many of you know you can give goals to your agents? You can just set a goal and forget about it. Anybody? Nice. Um so, what if you combine goals and loops? You can just set a goal saying there is some data discrepancy in this report. And in the production bug like the source data is not matching with what the agent has generated. So, you can just set a goal to fix this, look into this, set a loop. You can even close like close your laptop because you can do it from your phone nowadays. And if you look at the last but one point, which is dreaming, um let's say a lot of are using your production software and they are doing the same type of patterns and they're facing the same type of problems. So, you let the agent dream like humans dream in the background so that it collects all the sessions that your customers are using and compacts it into a set of data points which your system can use and basically \[snorts\] upgrade yourself. So, with a combination of all these features, basically you can essentially remove yourself out of the loop. But, as I said before, the developers do a lot of variety things in their software development life cycle and sitting behind a desk from 9:00 to 5:00 and just writing code is not valid anymore. So, just an overview of what I've covered till now in the session. You can have a team of sub agents working in parallel work trees. You can have skills, your organizational secret recipes, your customers recipes. You can give all of these to an agent. Your agent can connect to whatever third-party server there is. It can be a logging system, it can be an authentication gateway. And you just compress all of this into one pane of glass because minimal UX is the key. And you can set goals and loops for autonomy. If you think this particular work can be done by the agent a lot better, you can just ship it to the agent. Always have the human as a verifier, but not the throughput ceiling because human attention is very limited. So, thank you for your time and thank you for your Thank you for I hope you um learned something from the session. Thank you. \>\> \[applause\] \>\> Hey.