BitBoard Teaches AI Agents to Build Dashboards You Actually Get to Keep

Ask a modern AI to analyze your data and it will oblige in seconds. It will write the SQL, clean the numbers, draw the chart, and explain what it means. Then you close the tab, and all of it disappears. The query, the connection, the reasoning, the picture - gone. Next month you ask again, and the agent starts from nothing. BitBoard, a two-person company out of Y Combinator's Spring 2025 batch, was built on that small, maddening observation: the value was never the chart. It was the work that made the chart, and nobody was keeping it.

The live product today is straightforward to describe. "Dashboards built with your favorite AI tools," reads the homepage. You connect your data, you build with your agent - Claude, ChatGPT, or Cursor - and you share the result with your team. Instead of a one-off conversation, BitBoard hands back a hosted dashboard, a live report, or a workbook that holds the SQL, the Python, and the markdown behind it. The chat becomes an asset. The asset can be re-run.

01 / What it doesThe dashboard that remembers how it was made

Most analytics tools ask you to learn them. You open Tableau or Looker, you find the right menu, you drag the right field, and eventually a chart appears. BitBoard's bet is that the interface most people now reach for first is a chat window, and that the agent on the other side is perfectly capable of building the report - it just needs somewhere to put it. So BitBoard supplies the somewhere. It connects to your data sources over the MCP protocol, lets your agent push queries and results into a stored workbook, and hosts the output in a browser where colleagues can open it without a login ritual or a screen-share.

The word the company keeps returning to is traceability. Every query, every connection, every line of code sits inside the workbook, which means a dashboard is no longer a screenshot someone pasted into a deck three quarters ago with no way to check it. You can see what it asked, re-run it against fresh data, and hand it to a teammate who can do the same. In a field crowded with tools that make analysis faster, BitBoard is unusually focused on making analysis durable.

That distinction is easy to miss until it costs you. Anyone who has inherited a spreadsheet from a departed colleague knows the particular dread of a number with no lineage - a figure that everyone cites and no one can rebuild. BitBoard's answer is to treat the workbook, not the chart, as the deliverable. The chart is just the part that happens to be visible. Underneath it sits a documented trail that a new hire can open, follow, and trust, which turns a personal analysis into something the organization can actually own.

02 / The originIt began inside a healthcare startup that failed

BitBoard did not start here. When Connor Jones and Ambar Choudhury launched the company in 2025, the pitch was different: AI agents for healthcare back-offices. The two had met at Forward, a heavily funded primary-care startup where Jones worked on the business side and Choudhury was the first engineer. Forward shut down in November 2024. Before it did, the founders had spent their days coordinating thousands of remote operations contractors - the people who fill in intake forms, chase referrals, update patient charts, and triage the endless stream of faxes that still runs American healthcare.

That experience became the first product. BitBoard sold AI "workers" that behaved like remote contractors inside a clinic's existing systems. You showed the software your process, it trained an agent on that process, and the agent then worked quietly in the background - reading PDFs, filling forms, moving tasks through the EHR - without anyone having to learn a new interface. It was HIPAA compliant, priced per task by complexity, and wrapped in audit logs and deterministic verification checks so a clinic could trust that the work was actually done.

1,300+ hours

Recovered per customer, per year — original healthcare product

The numbers were the kind that make a founder feel like they have found the thing: more than 1,300 hours recovered per customer per year, and a live agent in about nine days from the first process review. But healthcare is a slow market to sell into, thick with compliance and long procurement cycles, and the founders kept noticing that the hardest part of every deployment was not the automation. It was the data - understanding a clinic's operations well enough to build anything at all, and keeping that understanding somewhere it would not evaporate.

03 / The pivotFrom AI workers to agent-built analytics

So BitBoard changed shape. The same instinct - let an AI agent do the repetitive work, then keep the result so nobody redoes it - pointed at a broader and less regulated problem. Everyone with a database and an AI subscription was now generating throwaway analysis. The company reoriented around that, and the name still fit both eras: a board of AI workers, or a board of dashboards. Today the front door is the analytics product, and the healthcare chapter reads as the origin story rather than the mission.

Pivots are easy to narrate after the fact and hard to do while a company is small. What is worth noticing is the through-line. Both products start from the same premise - that AI is good at the doing and bad at the remembering - and both try to fix the remembering. The healthcare version stored a clinic's process so an agent could repeat it. The analytics version stores an agent's analysis so a team can repeat it. The customer changed. The idea did not.

04 / Who it's for & how it chargesFree for one, flat for many

The target user now is anyone who already lives in an AI agent and touches data: analysts, operators, founders, and the growing population of people who would rather ask ChatGPT for a chart than open a BI tool. BitBoard meets them where they are. The individual plan is free, with unlimited access, free hosting, and no credit card - a deliberate choice to let people build first and decide later. Teams pay custom, flat-rate pricing for centralized admin controls, shared billing, and advanced security.

The absence of usage metering is a quiet statement of intent. Plenty of data tools charge by the query or the seat in a way that makes people ration their own curiosity. BitBoard's flat pricing is a bet that the way to win the agent-analytics market is to remove the meter and let usage compound. Whether that holds as the company grows is an open question, but it is a coherent one.

05 / How it's differentBuilt from the agent out

Most tools calling themselves AI-powered took an existing dashboard and bolted a chatbot onto the side. BitBoard went the other direction. It started from the agent and asked what a workbook would look like if an AI built it and a human kept it. That ordering matters. Because BitBoard does not try to be the intelligence - it connects to Claude, ChatGPT, and Cursor rather than shipping its own model - it inherits whatever the frontier labs release next. Betting on the MCP protocol instead of a pile of custom integrations is how a team of two can plug into tools that thousands of engineers are improving on their behalf.

Against classic business intelligence, the pitch is speed and reproducibility without a new query language to learn. Against the raw chat window, the pitch is memory and sharing. Against the AI-native notebook crowd, the pitch is that BitBoard is agent-first by default rather than a notebook that recently grew a copilot. None of these competitors is standing still, and BitBoard is very small. But the position is legible, which is more than many young analytics companies can say.

There is also a timing argument buried in the strategy. Every few months the underlying models get materially better at reading a schema, writing correct SQL, and reasoning about what a business actually wants to know. A tool that ships its own model has to keep pace with that on its own budget. A tool that connects to whatever model you already pay for gets the upgrade for free. BitBoard is structured to ride that curve rather than compete with it, which is a sensible posture for a company that cannot outspend anyone.

06 / The peopleTwo operators, one failed startup, no wasted lessons

Connor Jones, the CEO, came to Forward by way of Columbia Engineering and BlackRock - a finance-and-engineering background that shows up in the company's emphasis on audit trails and reproducibility. Ambar Choudhury, the CTO, was the first engineer at Forward and spent time at Palantir before that, a company whose entire reputation is built on getting messy operational data into a usable shape. Between them they had run large distributed teams and shipped software into one of the most demanding environments there is, and then watched the company around them close. BitBoard is what they did with that.

There is a version of this story where two people leave a failed startup and build a monument to the thing that failed. BitBoard is the other version. The founders took the specific, unglamorous lesson - most data work is redone, not done - and kept following it even when it led them out of the industry they knew. That willingness to change the product while holding onto the insight is, more than any single feature, the thing that distinguishes the company.

For now BitBoard remains a team of two, backed by Y Combinator with a modest seed and a product that is free to try and easy to explain. The market it is walking into is enormous and contested, and none of the incumbents will hand over their dashboards without a fight. But the company is aimed at a real and slightly embarrassing gap in how people use AI today: we generate brilliant, disposable answers all day long, and then throw nearly all of them away. BitBoard's whole proposition is picking them back up.