Jun 10, 2025 · 33m · a16z
Giving New Life to Unstructured Data with LLMs and Agents
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this a16z podcast episode, Instabase Founder and CEO Anant Bhardwaj joins Partner Guido Appenzeller to discuss how compound AI systems, compile-time agentic workflows, and federated execution frameworks are solving the challenge of unstructured data and transforming enterprise automation beyond legacy RPA.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The host as informed peer 3.6 Guest teaching 4.0 Guest disagreement 1.1 The host pushing back 0.6
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 13:49 Guest reframes accuracy vs predictability
Anant directly reframes Guido's question about error rate benchmarks, stating that enterprises do not care about raw accuracy as much as error predictability.
Hardest push from the host ▶ 31:34 Host challenges full identity pass-through for agentsGuido refuses the premise that agents should hold identical access rights to human users, using an intern spending limit analogy to advocate for strict privilege boundaries.
Biggest teaching moment ▶ 8:34 Guest breaks down failure modes of naive RAG on enterprise documentsAnant educates on why context windows and vector retrieval fail on complex financial packets due to missed table cells and lack of completeness guarantees.
The host holds their own ▶ 32:35 Host frames executive strategy using historical tech cyclesGuido takes control of the segment to deliver strategic guidance to enterprise leaders, drawing historical parallels to the dot-com revolution and Barnes & Noble.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| Title Card: LLMs and Agents for Unstructured Data | 3 | 5 | 1 | 0 | Anant details the evolution of unstructured data processing from MIT research to Instabase, BERT, and InstaLM. Guido demonstrates tech familiarity by mentioning two-dimensional rotary encoding and the bitter lesson. | |
| Enterprise Use Cases and System Reliability Beyond LLMs | 3 | 5 | 1 | 0 | Guido shares a personal anecdote about sorting scanned PDFs before Anant explains why standard RAG and LLMs fail on complex financial documents due to precision versus completeness trade-offs. | |
| Predictability vs. Perfection and User Experience Shifts | 4 | 4 | 2 | 1 | Guido asks how enterprise compliance acceptance criteria are shifting from absolute perfection to human error benchmarks. Anant gently reframes the issue, explaining that predictability of errors matters more than raw accuracy. | |
| Conversational Lending and Barriers to Enterprise AI Adoption | 3 | 4 | 1 | 1 | Anant shares examples of conversational lending over WhatsApp and notes enterprise slowness. Guido offers a mild counterpoint that enterprises are moving faster with AI than in past tech cycles. | |
| AI Agents in the Enterprise: Compile-Time vs. Run-Time | 4 | 4 | 1 | 0 | Anant introduces the distinction between build-time compile-time agentic generation and deterministic runtime execution. Guido synthesizes this with broader industry debates on full autonomy versus workflow freezing. | |
| Future Vision: Federated AI Execution Frameworks | 2 | 5 | 1 | 0 | Anant shares his long-term vision of federated AI execution frameworks replacing traditional robotic process automation (RPA) in enterprises. | |
| End-to-End Workflow Automation and Model Context Protocol | 5 | 4 | 2 | 3 | Anant outlines how Model Context Protocol and identity pass-through enable end-to-end workflow automation. Guido pushes back on full capability pass-through, arguing agent privileges should be capped like an intern. | |
| Conclusion and Strategic Lessons for Enterprise Leaders | 5 | 1 | 0 | 0 | Guido synthesizes strategic lessons for executives, comparing AI adoption to the dot-com era where delay risked obsolescence. Anant summarizes three core business outcomes. |
Statements from this episode (13)
Bhardwaj defines unstructured data as anything non-queryable via SQL
“Anything that cannot be put into nice database tables where you can run SQL, anything that is not that is unstructured data.”
Instabase tripled its revenue between 2021 and 2022
“We tripled our revenue that year, twenty-twenty-one to twenty-twenty-two.”
Bhardwaj: Automated systems process bank lending applications in under five seconds
“So now you can do lending in like less than five seconds. Rather than earlier, that would have taken several, several weeks.”
Bhardwaj: RAG alone is insufficient for high-accuracy enterprise data processing
“While drag is good for casual search, you need a complex workflow under the herd that is explainable, that is auditable, that is guaranteed to be accurate and correct, is important for solving many of these enterprise problems.”
Bhardwaj: Major capital will flow into building reliability systems around LLMs
“And that is going to be a lot of investment that you will see across the board, which is how do we build the right systems around AI and LLMs that solves the problem.”
Bhardwaj: Enterprise AI clients prioritize predictability over baseline accuracy
“So I think in general what we have seen is enterprises are fine using AIs as long as we show them predictability. They don't care about, you know, 99% accuracy. You can be 90% accurate or even 80% accurate, But just tell us which 20% need to be reviewed or whi…”
Bhardwaj: AI will automate filtering unstructured documents down to key reviewable items
“Whenever unstructured data like documents come in, humans will still see some kind of dashboard with like whatever stuff is, and only the thing of interest they will go and double click on. And AI will do a lot of things to minimize their time to get to that t…”
Bhardwaj: Indian bank conducts entire lending process conversationally over WhatsApp
“I was working with a bank in India, and now given that AI is, has become reasonably reliable, they are offering entire lending over WhatsApp.”
Bhardwaj: Enterprise AI adoption hinges on data security and auditability
“Two key things that they care about is, how do you guarantee that my data is safe and secure? So that's number one. And second is, how do you give me auditability and predictability? That's the two more, like if you boil down to all their questions, they event…”
Bhardwaj: Enterprises reject black-box AI decisions without clear step-by-step explainability
“Nobody wants, like, AI making a decision, even if it is correct, if they cannot Explain here are the set of steps that it took, because if something wrong happened, they have to explain, because in human world you can explain.”
Bhardwaj: Enterprise AI agents will operate at compile-time, not runtime
“So I do not believe that autonomous agent would be a runtime phenomena. However, there would be a build time or compile time phenomena.”