Future of Work · 5-minute read
A faster review is valuable only when a qualified person can explain what the system found, what it missed and why the conclusion should be trusted.
Legora used GPT-6 Astra to review 41 documents in minutes as part of a financial-statement workflow. It is an impressive demonstration of professional-scale context and tool use, but it does not erase expertise; it changes where expertise must be applied.
What happened
Legora, an agentic system for legal and professional work, tested Astra on a document-heavy financial-statement review. OpenAI’s customer material says the system analysed 41 documents in minutes and improved on Legora’s internal benchmark for the task.
Astra’s technical profile helps explain the result. The API model supports a context window of 1.05 million tokens and tools for file search, code execution and structured outputs. Those capabilities allow an agent to collect evidence across many files, perform calculations and return a formatted result without treating each document as a separate conversation.
The meaningful advance is workflow continuity. The model is being asked to maintain an objective across a body of evidence, not merely summarise one contract at a time.
Why it matters now
Professional services are built around attention. Lawyers, accountants and analysts spend large amounts of time locating clauses, reconciling figures and moving evidence into standard formats. Automating that connective labour can return time to interpretation, negotiation and client judgement.
It can also change the economics of expertise. If the first review becomes dramatically faster, clients may expect quicker answers and lower fees. Firms may process more material with smaller teams. Junior professionals may encounter fewer of the repetitive tasks through which they once learned how documents fit together.
The profession therefore needs a new training loop: people must learn to interrogate model-produced work without losing the underlying craft required to detect a confident mistake.
What changes
- For professionals: Review moves from locating every fact manually towards designing checks, resolving ambiguity and accepting responsibility.
- For firms: Audit trails, source citations and approval gates become part of the service clients are buying.
- For clients: Faster delivery should create an expectation of transparency, not an assumption that the model is infallible.
The tension
A system can appear authoritative because it has read more material than any individual could read in the available time. Yet professional accountability cannot be delegated to context-window size. A missed exception, an outdated rule or a subtle conflict between documents can still change the answer.
The Agentica IX view: Astra does not make professional expertise obsolete. It makes unsupported expertise harder to justify. The strongest practitioners will combine faster evidence gathering with clearer reasoning, visible sources and an explicit human decision.
What to watch next
- Whether professional regulators define standards for agent-assisted review.
- How firms train junior staff when repetitive document work declines.
- Whether clients receive usable evidence trails rather than polished conclusions alone.
Sources: OpenAI: Legora financial-statement review with Astra and OpenAI API: GPT-6 Astra model and OpenAI: GPT-6 Astra launch.