A Legal Firm in Your Inbox – for the Cost of a Tool Subscription

01 /The Concept

Every business receives contracts from counterparties. Reviewing them properly requires a lawyer, which is expensive and slow. Crosby collapses that bottleneck – an AI legal firm that assesses whether an incoming contract is safe to sign as-is, or whether changes are needed to avoid unfavorable terms.

Clients submit documents to Crosby directly via messenger, email, or automatically by API – for example, by routing all incoming counterparty contracts to Crosby for pre-review before anyone on the legal or business team looks at them.

The submitted contract first goes through Crosby's AI engine, which flags questionable clauses, proposes specific edits and additions, and drafts a reply letter to the counterparty that enumerates the requested changes with reasoning.

Human lawyers at Crosby then review and refine the AI's work – checking that nothing was missed and that the AI responded correctly to any subtle provisions buried in the text.

This combined output from AI and human review is what Crosby delivers to the client.

Unlike traditional law firms, Crosby doesn't bill by the hour. Each document has a fixed price that clients know upfront before submitting, with pricing determined by document type and length.

The guaranteed turnaround is four hours from receipt, though most results arrive in minutes. That speed, combined with predictable pricing, is a decisive advantage over any firm working by traditional manual methods.

Importantly, Crosby's AI reviews more than legal terms – it also checks the numbers embedded in a contract, such as late-delivery penalty amounts, and benchmarks them against market standards using data drawn from a continuously updated database of real executed agreements.

The AI also builds a proprietary knowledge base for each client based on the contracts it has processed for them over time – adapting its recommendations to the client's specific context and deal patterns.

Crosby is still in closed beta but its metrics are moving fast. It took 173 days to process the first 1,000 contracts; it now processes 1,000 contracts every three weeks. The client count is growing 30% month over month.

Crosby was [covered previously](/review/dlja-odnih-jeto-povod-dlja-rasstrojstva-a-dlja-tebja-sposob-zarabotat) when it raised its initial $5.8 million in June. Less than four months later, it has closed a new $20 million round – which is reason enough to revisit.

02 /Why It Matters

Crosby's business model is genuinely interesting – not just because someone had the idea of embedding AI inside a law firm, but because of what the AI is specifically constrained to do: review and improve existing contracts rather than draft them from scratch.

The key insight is that at the current stage of AI development, these models are far better at improving and extending what humans have already created than they are at inventing from nothing. The same pattern appears across several other startups worth noting.

Rocketable ([related review](/review/v-obshhem-sluchae-jeto-poka-fantastika-a-v-chastnom-vozmozhnost-na-milliard)) is building a software holding company designed to run with minimal human staff – relying on AI for almost all product work. But the AI isn't creating new products. Rocketable acquires already-functioning cloud services built by human developers, then hands them to AI for bug fixes, optimization, and feature expansion. The founder's explicit bet is that AI can develop but not yet originate. The startup graduated from Y Combinator this year with $6.5 million raised.

Convexia ([related review](/review/chtoby-sorvat-bolshoj-kush-nuzhno-stat-maksimalistom)), a September Y Combinator graduate, is building a platform to bring new drugs to market ten times faster by automating the development process with AI. But again: the AI doesn't invent the drugs. Instead, it searches research databases, lab publications, and patent registries for undervalued compounds already discovered by humans – then assesses market potential, identifies likely acquirers among major pharma companies, and prepares the most promising candidates for commercialization. Human experts still make the final selection calls, using the AI's analysis as their foundation.

And Then ([related review](/review/a-vot-jetim-malo-kto-zanimaetsja-nu-i-ochen-zrja)) applies the same logic to interactive storytelling. The startup, just out of a16z's Speedrun accelerator, builds voice-based interactive games – dialogues between users and an AI. In one game, the player is sitting in a 24-hour diner when a man walks in with a bomb. The founder believes AI can only produce formulaic plots; the story concept is always written by a human. But AI handles all the real-time dialogue with each player, branching uniquely based on responses – which is exactly what it does well.

03 /Opportunities

The broader direction: AI platforms where the AI doesn't originate but instead corrects, enhances, and develops what humans have already created.

This pattern can be applied in almost any field. A simple example from the startup world itself: a person describes a product idea, and an AI engine searches for competitors, analyzes their feature sets, and returns a prioritized list of ways to differentiate. The human creates the input; the AI does the analytical heavy lifting.

Following Crosby's full model, human experts would still review and verify the AI's output – which substantially increases the value of the platform. And the price.

Although startup analysis is probably not the specific vertical worth targeting – something simpler and more mass-market would be a better bet.

The most interesting entry points are mass-market domains with high error costs – where humans already produce a first draft and professionals currently charge a premium to review it. Tax prep, medical records, insurance claims, construction plans. Pick one, add AI that improves the draft and human experts who verify the output, and the model writes itself.