AT&T announced this week that it will deploy OpenAI chatbots and agents within its in-house legal team and cut its external law firm roster from 30 to 18–20 over two years. On its surface, this looks like the inevitable future: big corporates take commodity work in-house, AI handles the grunt, fewer external lawyers needed. But for UK mid-market law firms, insurers, accountancy practices and financial services firms reading this, it should signal something more important: this is what happens when a large enterprise with unlimited budget and internal IT resources tries to treat AI deployment as a simple vendor swap. AT&T will likely spend millions on integration, training, governance and remediation before it realizes that swapping external lawyers for OpenAI agents solves a cost problem but creates three compliance and liability problems. The question is whether your firm learns from AT&T's approach or from a different model entirely.

The pattern here is familiar. Large corporations are now using raw LLM capability (often via OpenAI or similar APIs) to attempt wholesale insourcing of legal work, document review and contract analysis. Harvey, Legora and Luminance have all positioned themselves as 'legal AI' solutions, and they have real strengths in document classification and pattern recognition. But the dirty truth none of them adequately address is that deploying a general-purpose LLM agent into a regulated business's legal function is not a cost-reduction play—it is a compliance and delegation play, and both require active governance, continuous audit and explicit regulatory alignment. The FCA's recent emphasis on governance frameworks (echoed in PRA SS1/23 and the ICO's UK GDPR guidance) means that using AI to reduce headcount without simultaneously building transparent, auditable decision-making infrastructure is now a regulatory liability, not a strategic win. AT&T's move signals that large corporates are betting they can absorb this cost. Mid-market firms cannot.

Here is Trovix's direct view: AI should augment in-house teams, not replace external judgment, and it should do so within a clear governance boundary. AT&T's approach—which appears to be 'deploy OpenAI agents and watch headcount fall'—conflates two very different things: reducing spend on routine work (legitimate) and removing human accountability for legal decisions (dangerous). The difference matters. A solution like Legora might excel at bulk document triage, but it does not tell you whether you are compliant with FCA Consumer Duty PS22/9 or the SRA Code when you deploy it. Microsoft Copilot for Enterprise offers better data residency and governance controls, but it still does not give you the audit trail a regulated firm needs if a claim arises. Trovix Audit exists precisely because we believe the future is not 'fewer lawyers with AI', it is 'same number of lawyers, working faster, with auditable AI decisions'. Fee-earners need knowledge assistance (which Trovix Aria provides via RAG architecture and source attribution), not replacement. Clients need a clear, compliant interface to your AI, which Trovix Reach delivers. The insourcing trend AT&T represents will fail, quietly, when the first audit finds that an AI agent made a decision without proper delegation or when the first claim alleges the firm outsourced judgment to a black box.

If you run a law firm, accountancy practice, insurer or financial services firm, do not let AT&T's announcement push you toward panic hiring or blind AI adoption. Instead, ask three hard questions of any AI deployment in your regulated business: (1) Does it create an auditable decision trail that would survive an FCA visit or PRA inspection? (2) Does it reduce work volume or does it reduce headcount—because those are not the same, and regulators care about the latter? (3) If a claim arises tomorrow and involves a decision made or assisted by this AI, can you explain, in writing, why a human kept genuine accountability? If your answer to any of these is 'no', you are following AT&T's path, not your own. The firms that will win in the next two years are not those that cut lawyers fastest. They are those that amplify lawyer judgment, cut wasted time and keep regulators confident that humans remain in charge.

Source: Crypto Briefing