John Corcoran

Fmr Clinton White House | recovering attorney | AI | podcaster since 2010 | EO Member

[John Corcoran @JohnCorcoran]

"but what if a car gets into an accident?"

human drivers kill 40K people in the U.S. every year

I can't wait until every car has an autopilot

[ΛI DRIVR @AIDRIVR]

FSD reaction time testing @ ~45 mph

@DavidMoss and I were determined to kill this dummy

[ΛI DRIVR @AIDRIVR]

FSD reaction time testing @ ~45 mph

@DavidMoss and I were determined to kill this dummy

[John Corcoran @JohnCorcoran]

Mark my words, within a year, SpaceX will buy Tesla.

[John Corcoran @JohnCorcoran]

@samdblond It's a cool idea, but you are absolutely not the first person to do aerial advertising over SF.

[John Corcoran @JohnCorcoran]

Legal ai also will decrease costs for a necessary cost of doing business which has been so expensive for vast swaths of the market that many have just done without. Which leads to more legal disputes and issues.

[Marc Andreessen 🇺🇸 @pmarca]

Legal AI superempowers normal individuals with no legal background to fight big institutions in bureaucracies and in courts on a level knowledge/skill playing field, for the first time in human history. As such, it is one of the most inspiring applications of AI.

[John Corcoran @JohnCorcoran]

@karpathy Wow this is huge news

[John Corcoran @JohnCorcoran]

Attorneys need to adapt quickly to these changes

[Ann Srivastava @helloparalegal]

A litigation attorney called me last month. Solo, mid-sized matter, opposing counsel was a partner at a 200-lawyer firm. Deposition the next morning. She was drowning.

"I have 14 deposition transcripts, 9 expert reports, 200+ exhibits. I can't keep track of who said what. The associate I'm using is six months out of school and is missing things."

I asked her two questions:

- How are you organizing this right now?
- Have you ever heard of Obsidian?
She was using folders. Word docs. Yellow stickies on her monitor. She had not heard of Obsidian.

I told her to do one thing before the deposition.

Set up an Obsidian vault. Drop every transcript, every exhibit summary, every witness profile in. Plain markdown files. Backlink every name across the matter. Run one Claude Code query against the vault before she walked into court.

The query: "Read every deposition in this matter. Identify any place where this witness contradicted earlier testimony, or where any other witness contradicted what this witness said. Quote the exact text."

She set it up Saturday afternoon. Four hours. Free.

She ran the query Saturday evening.

Three contradictions came back. One was material.

In the witness's own deposition six months earlier, he had testified that he had not been in the warehouse on the night in question. In a 30(b)(6) deposition three weeks later, the company's HR director produced a swipe-card log placing him there. Her junior associate had reviewed both depositions separately and never connected them.

The agent connected them in 40 seconds.

Sunday she walked into her co-counsel's office and dropped the printout on his desk.

"Where did you get this?"

"From a markdown file."

The Monday deposition went very differently than expected.

She is not the only litigator I have helped do this in the last six months. The pattern is always the same.

The lawyers running their cases out of folders, Word docs, Clio "matter notes," and human memory are leaving facts on the table.

The ones running their cases out of Obsidian + an API-direct agent are finding the things their associates miss.

Free plugins do the structural work: Templater for new-matter intake, Dataview for deadline dashboards, Tasks for court deadlines.

The agent layer (Commercial Terms API, ~$50/mo) does what no SaaS will ever do for you:
→ Cross-matter witness lookup. "What do I know about Expert Foster across all matters?"

→ Deposition outline drafting. Reads the witness profile, matter facts, your prior outlines.

→ Exhibit summaries. New PDF lands in Discovery, summary appears the next morning.

→ Timeline construction. Reads every transcript in a matter, builds the chronology.

→ Pattern cross-reference. "Has opposing counsel run this argument before in another matter?"

Most litigators are paying $500-900/mo per attorney for SaaS that does none of this. Clio does not do cross-matter pattern matching. Westlaw does not read your own depositions. Harvey is a chatbot with the agent features turned off.

The "workflow you own" isn't a slogan. It's a directory structure plus an API key.

Most litigators will not build this. The ones who do will keep finding the things their associates missed in 40 seconds, on a Saturday evening, before a Monday deposition.

This is what "own your workflow" actually looks like for a litigator. Here is the structure:

[John Corcoran @JohnCorcoran]

A couple of years from now, anyone who uses a keyboard or logs directly into their apps will look like someone using a typewriter or an abacus.

[John Corcoran @JohnCorcoran]

Out: fingers on keyboards.

In: whispr flow

Out: working directly inside of apps

In: directing your agent swarm to act on your behalf

[John Corcoran @JohnCorcoran]

@AlexFinn I have Tesla solar and OpenClaw, and I like the idea of it, but solar rarely covers it all your power needs

[John Corcoran @JohnCorcoran]

@garrytan Glad you are safe. So sad that it ended that way, but glad no one else was really majorly injured.

[John Corcoran @JohnCorcoran]

@Jason I’m in

[John Corcoran @JohnCorcoran]

Going to pick up my son from soccer; Claude, you take it from here

[John Corcoran @JohnCorcoran]

@SMB_Attorney And I’ll bet most attorneys could get by on the $20/mo plan

[JNS @_devJNS]

me and Claude building an app at 3AM

[John Corcoran @JohnCorcoran]

WILD

[Garry Tan @garrytan]

OpenClaw is the most significant innovation since ChatGPT and Claude Code

It’s the third shockwave that will be felt by all

[John Corcoran @JohnCorcoran]

Don't give up. It's been nine years, and he is a fifth of the way there.

[Carter Wilkerson @carterjwm]

HELP ME PLEASE. A MAN NEEDS HIS NUGGS

[John Corcoran @JohnCorcoran]

Out: fingers on keyboards.

In: whispr flow

Out: working directly inside of apps

In: directing your agent swarm to act on your behalf

[John Corcoran @JohnCorcoran]

@tenobrus @garrytan I'm a Wispr Flow super user (400k+ words dictated). I fail to see it. Especially when it comes to actually whispering your dictation, Which is super useful at times. Wispr Flow definitely beats Aqua in that category.