Cohere Compass 2 vs Harvey AI Due Diligence Agent

“The primitive here is a retrieval pipeline that can ingest mixed-format documents — PDFs with embedded charts, scanned tables, image-heavy slides — and return semantically relevant chunks without requiring a preprocessing ETL step per modality. That's a real problem: anyone who's tried to build RAG over a 10,000-document enterprise library knows the pain is 80% in the ingestion layer. The DX bet is that Cohere handles the multimodal parsing so you don't glue together a PDF parser, a table extractor, and a vision model yourself. The on-prem deployment option is actually the headline feature for the buyer, not the multimodal part — that's what gets it past legal review. My skip concern is documentation: the blog post is long on capability claims and short on API surface, schema design, and what 'image understanding' means at query time versus index time. Show me the query API, then we'll talk.”

“The primitive here is: document ingestion pipeline plus structured extraction plus risk taxonomy, wrapped in a workflow UI. That's legitimate engineering — OCR normalization, citation grounding, and hallucination mitigation on legal text are genuinely hard problems. But I can't evaluate the DX because there is no public API, no developer documentation, no SDK, and no pricing I can read without talking to a sales rep. The blog post is marketing copy with a screenshot. If this is purely an enterprise workflow product that lives in a GUI, fine — but the review stops at the door because there's nothing to verify. Ship when Harvey publishes an API reference or at minimum a technical architecture post; skip on the current evidence because 'trust us, it works' is not a technical decision I can recommend.”

“The direct competitors are Azure AI Search with multimodal indexing, AWS Kendra, and increasingly any RAG stack bolted onto GPT-4o's native PDF vision. Compass 2's real differentiator is not the multimodal capability — every major cloud provider is shipping that — it's the on-premises deployment for enterprises with data residency requirements, combined with a retrieval model trained specifically for enterprise document retrieval rather than general web content. The scenario where this breaks is at the 'chart understanding' claim: interpreting a bar chart semantically in a way that survives a specific quantitative query ('find all documents where Q3 revenue exceeded Q2') is a much harder problem than the blog post implies, and I've seen this class of tool hallucinate chart data confidently. What kills this in 12 months isn't a competitor — it's that the chart and table comprehension doesn't hold up under production query loads and the feature gets quietly deprioritized. I'm shipping it narrowly: for text-heavy PDFs with some visual elements in air-gapped environments, this is probably the best available option right now.”

“Harvey is doing something genuinely harder than most legal AI: not just answering questions about documents but running an end-to-end workflow across an unstructured data room and producing a structured issue list that a lawyer would actually hand to a client. The direct competitor here isn't ChatGPT with a custom prompt — it's Kira Systems, Luminance, and Relativity, all of which have years of training data on deal documents. Harvey's bet is that frontier model quality plus legal-specific fine-tuning beats purpose-built classifiers, and for nuanced contract interpretation that bet is probably right in 2026. What kills this in 18 months: if Anthropic or OpenAI ships document-native reasoning APIs good enough that any firm's IT team can stand up a comparable workflow, Harvey's moat shrinks to go-to-market and training data — which is real, but thinner than it looks.”

“The buyer is a VP of IT or Chief Data Officer at a regulated enterprise — financial services, pharma, government — and the budget comes from the data infrastructure or compliance line, not a software tools budget. That's a real check-writer with a real problem: they have document libraries they legally cannot send to OpenAI's API, and they need search that works across formats. The on-prem deployment option is the actual moat here, not the multimodal capability — Cohere has been building that distribution channel for two years and it creates genuine switching costs once it's integrated into an enterprise's document management stack. The risk is that the pricing model is 'contact sales' all the way down, which means a long sales cycle and high CAC that has to be recovered on large contracts. What survives the model-gets-cheaper scenario is the enterprise integration layer and compliance certifications, not the retrieval model itself — Cohere needs to be pricing for that, not for compute.”

“The buyer here is the AmLaw 200 firm or the Big Four legal department, and this comes out of deal advisory budgets that routinely run seven figures per transaction — Harvey's pricing is a rounding error against that backdrop, which is the correct place to anchor. The moat is real and layered: enterprise data room integrations are sticky, associates trained on Harvey outputs don't go back, and the feedback loop from reviewed deals compounds into training data competitors can't replicate. The risk isn't pricing pressure, it's scope — M&A due diligence is episodic revenue, not recurring, and Harvey needs to colonize the ongoing contract management and regulatory review workflows to build the expansion story. They know this; the question is execution speed before well-funded competitors like Ironclad and Lexion expand upmarket.”

“The thesis Compass 2 is betting on: enterprise knowledge is fundamentally multimodal — it lives in slide decks, scanned contracts, financial tables, and annotated diagrams — and the first retrieval system that treats those formats as first-class citizens rather than edge cases will own the enterprise search layer. That's a plausible and falsifiable bet, but the dependency is that 'understanding' a chart means something semantically useful at query time, not just 'we embedded the image.' The second-order effect that matters here isn't faster document search — it's that if this works, structured data that currently lives locked in PDFs becomes queryable without a data engineering team to extract it, which shifts power from BI teams who own structured pipelines toward anyone with a document library. Cohere is riding the trend of on-premises LLM deployment for regulated industries — that trend is real and accelerating, and they're on-time to it, not early. The future state where this is infrastructure is 'every regulated enterprise has a Compass instance the same way they have an Active Directory instance.' I'd believe that in five years if the chart comprehension claim is real.”

“The thesis here is falsifiable: by 2028, the bottleneck in M&A deal timelines shifts from lawyer availability to data room quality, because autonomous agents can absorb document volume that would have required a 40-person associate team. That's not a vibe — it's a specific claim about where deal friction lives, and it's directionally correct given current associate billing rates and deal timeline compression pressure. The second-order effect that nobody is talking about: if Harvey normalizes autonomous issue list generation, the junior associate due diligence role hollows out faster than law school enrollment adjusts, and firms that adopt early capture margin that was previously paid out in associate salaries. Harvey is on-time to this trend — not early, not late. The infrastructure state where this wins is Harvey becoming the default data room intelligence layer, the way Kira was for contract review before LLMs made Kira's classifier approach look dated.”