Colin Ashford

Take us back. What did a litigation week look like before you brought AI into it?

Discovery is where the hours went. In a mid-size commercial dispute you can be handed tens of thousands of pages and a stack of deposition transcripts, and someone has to actually read all of it. For years that someone was me or a junior associate, working nights to build a chronology and pull the documents that mattered. A single day of deposition testimony might run three hundred pages, and turning it into something usable took most of a day.

The work was necessary, but it was slow, and it was easy to lose the thread of the case while buried in the volume. I spent more time locating facts than thinking about what to do with them. That imbalance is the thing I most wanted to fix.

What was the first thing you tried, and how did it go?

The first real test was a deposition summary. I fed a long transcript into a legal AI tool and asked for a narrative summary with page and line citations for every point. It came back in a few minutes with something I could work from, and the page-line references let me jump straight to the source and confirm each line. That mattered, because my instinct was to distrust it. The first draft was not perfect. It flattened some testimony a human reader would have flagged as evasive, and it missed a subtle contradiction between two answers. But it got me most of the way to a working summary in minutes instead of hours, and the citations meant I could check its work rather than take it on faith. That was enough to keep going.

How do you actually use these tools now, in your own routine?

Three ways, mostly. I summarize deposition and hearing transcripts, with citations, so I can compare testimony across witnesses and spot inconsistencies. I build case chronologies from the document set, which used to be a manual timeline in a spreadsheet. And I use retrieval tools to surface the key documents in a large production, asking questions in plain language instead of running keyword searches and hoping.

I also draft first-pass discovery responses, objections and answers, that I then rewrite. The pattern is always the same: the tool does the first read and the first draft, and I do the judgment. It handles the volume, I handle the strategy. On a typical week that gives me back close to ten hours I used to spend on manual review, and I move those hours to argument and client counsel.

Where do you not trust it, and where do you keep control?

Anywhere it touches the record or the court. I do not file anything a tool wrote without reading every line myself, and I verify every legal citation against the actual authority, because AI can produce case cites that look right and do not exist. Attorneys have been sanctioned for filing hallucinated cases, and the rule is simple: I am responsible for what goes out under my name, regardless of what drafted it.

Confidentiality is the other hard line. I only use tools with terms that keep client data private and out of training, and I think carefully before putting sensitive material into any system. My duty of candor to the court and my duty to the client do not delegate to software. The tools give me a faster first draft and a faster first read. The filing, the judgment and the accountability stay with me.

Was there a moment it clearly earned its place?

A contract dispute with seven depositions and a production in the tens of thousands of pages. Opposing counsel's client had given testimony across three sessions that I suspected did not line up. I ran all three transcripts through summaries organized by issue, put the witness's answers side by side, and found a direct contradiction on when a key decision was made. Doing that by hand would have taken days of reading, and I might not have caught it while managing everything else on the matter. With the summaries in front of me, it took an afternoon. That contradiction became a real point on cross-examination and changed how the other side valued the case. The tool did not find the argument for me. It put the raw material in a shape where I could see the argument. That is the distinction I keep coming back to.

What would you say to a litigator who is skeptical of all this?

I understand the skepticism, and some of it is warranted. My advice is to treat these tools as a very fast first-year associate, not as an oracle. You would never file an associate's memo without reading it, and you would never assume their case cites are good without checking. Apply the same discipline here and the risk drops sharply. Start with something low-stakes and verifiable, like a deposition summary where every point carries a page-line cite you can confirm in seconds. Do not put confidential material into a consumer tool. When a peer tells me they tried it once, got a bad answer and quit, my response is that they skipped the verification step that makes it safe and useful. The lawyers getting burned are the ones who trusted the output. The ones getting value are the ones who check it.

How has this changed your work, and what your clients get?

The clearest change is where my time goes. I spend far less of it locating facts and more of it thinking about the case, which is what clients are paying a litigator to do. Summaries that took most of a day take minutes, and I get roughly ten hours back in a typical week. For clients that shows up as faster answers, more thorough preparation on the same budget, and fewer hours billed for mechanical review.

It also changes what I can catch. Scanning an entire production instead of a sample means fewer important documents slip past. None of this changes who is accountable. Every filing is still mine, reviewed and signed by me. What has changed is that the grinding first pass is no longer the bottleneck, so the strategic work that actually decides cases gets more of my attention.