Most of the AI conversation in legal right now is about speed. Do the same work faster, bill less, move on. That framing misses something important. If the underlying process is mediocre, faster just means mediocre at scale. The more interesting question is whether AI can actually change the quality of the decisions we make on a deal.
This webinar was hosted by Laura Frederick of How to Contract and featured Justin Daniels, Shareholder and Corporate M&A and Tech Transactions Attorney at Baker Donelson. Justin has been building his own AI agents inside his firm's approved tooling rather than just using off-the-shelf chatbots, and he pulled back the curtain on how he designs them, what they produce, and where he pushes back when they get things wrong.
The conversation covered the four-step design process behind a useful legal AI agent, how to use AI to read counterparty personality and motivation, why arguing with the AI matters more than accepting its output, and how a disclosure schedule playbook for sellers can free up real attorney time for higher-value work.
Here are our top ten takeaways from the speakers' comments during the webinar:
Stop chasing speed and start chasing decision quality. Everyone is asking how to do legal work faster with AI. That is the wrong question. As Justin put it, if your process is mediocre, AI just makes you mediocre faster. The better frame is how we think better and negotiate smarter, because the real value sits in decision quality, not minutes saved. That reframe also changes how we talk to clients about AI value, since the billable hour does not capture it.
Design the workflow before you build the agent. Map the full process end to end before you write a single prompt. Justin said retrofitting a missed step is far more painful than thinking it through up front. The workflow design lets us see gaps in our own thinking before the tool ever runs, which usually saves us from building something with a fundamental flaw we only catch later when we are mid-deal.
Visualize the output you want before you prompt it. Picture the chart, the table, or the briefing memo in your head, then translate that picture into the prompt. Justin's negotiation agent produces a table with our position, their position, anticipated counterparty arguments, and risk level for each issue, because that is what he wanted to see. We get better output when we tell the AI what good looks like instead of letting it default to a generic format.
Keep a human in the loop at every meaningful step. Justin built explicit pause points into his agent where it asks the lawyer to review, add missed issues, and approve before the next step runs. Competent legal practice has to be done by a human, and AI tools should be treated as smart but inexperienced colleagues. The advice we give the client still goes through our bar license, no matter what tool helped us get there.
Use AI to read the counterparty, not just the contract. Justin loaded emails and documents into his agent to surface what kind of negotiator he was dealing with, whether the relationship-first type, the analytical type, or the hard bargainer. He also uses it to dig out motivations that are not on the surface. On one deal, the analysis showed his client was being hired so the counterparty could blame them later if regulators pushed back, and he advised against doing the deal even after winning every term.
Tailor the negotiation script to the personality type. Once we know who is across the table, the script should change. For an analytical counterparty, lead with facts and structure, give them silence to process, and avoid emotional pressure. Justin's agent produces opening moves, strategic questions, and reframes designed for that specific type, so we walk in with the language already prepared instead of inventing it on the call.
Defuse the elephant in the room BEFORE you start talking. When the client has handed off a deal to us at the last minute and we have to raise new issues with a counterparty who thought the deal was done, the worst move is to dive into the substance. Justin opens those calls by acknowledging the frustration directly, something like "I bet you got on this call thinking the deal was done and I'm sure that's frustrating." That one sentence lowers the other side's guard before the real conversation starts.
Argue with the AI instead of accepting what it gives you. AI pattern-matches on language and will read deal-specific facts through generic templates that do not fit. Justin had an agent congratulate a counterparty for a "big concession" that was no concession at all because they had anchored high to start. Push back. The argument is where the agent's weaknesses show up, and it is also how the prompts get sharper. Don't accept pattern matching. If the advice feels generic, it probably is.
Treat AI as a probabilistic machine that is trained to be helpful. That combination means it can be confidently wrong, especially on points where a small wording difference carries real legal weight. Justin's example was an agent calling a carve-out of negligence from the limitation of liability "reasonable" in professional services, when the gap between negligence and gross negligence is enormous in litigation. Our BS detector has to stay on.
Use AI to take rote work off your plate so judgment work gets your attention. Justin built a disclosure schedule guide that translates each rep and warranty into plain English for first-time sellers, with a box under each one for them to fill in. That tool produces in ten minutes what used to take hours of attorney handholding, and the seller's schedules come back cleaner. The mental capacity we get back goes into the negotiation and the bespoke drafting where our experience actually moves the deal.
Subscribe to Stay in the Loop
Our weekly newsletter brings you future webinar invites and recaps like this one. Subscribe now so you catch the next conversation even if you can't make it live.
