Articles
Mar 2, 2026
4 min read
4 min read

AT&T Just Built an In-House Legal Agency. Here's What to Steal (No OpenAI Deal Required).

Six lessons from AT&T's LegalEdge and BMW's legal AI rollout that any in-house legal team can apply, no big AI partnership required.

AT&T Just Built an In-House Legal Agency. Here's What to Steal (No OpenAI Deal Required).

Last week AT&T announced that its in-house legal agency, LegalEdge, is now a "flagship in-house legal design partner" with OpenAI. On the same day, Legal IT Insider reported that BMW Group is rolling out Legora across its legal department. Two global companies, two very different vendors, and one shared message: big legal departments have stopped "exploring AI" and started redesigning how work moves.

Before you forward the press release to your CFO with a hopeful "can we do this?", here's the good news. Most of what makes these programs work has nothing to do with having a famous AI lab on speed dial. Here's what to steal.

What actually happened

According to AT&T's October 8 announcement, LegalEdge launched in January 2026 as the department's "AI-first center of expertise" and handled nearly 100 matters in its first six months. With OpenAI, AT&T says it will design, build, and evaluate legal workflows with attorney oversight. The goals: make institutional knowledge easier to find, handle routine work faster, and free lawyers for strategy and advocacy. Artificial Lawyer described LegalEdge as a lean, cross-functional team of attorneys, paralegals, technologists, and legal ops and eDiscovery specialists.

BMW's rollout, per Legal IT Insider, covers legal research, document review, and drafting. First drafts are based on approved internal standards and precedents. The tool works inside Microsoft Word, and adoption is backed by a structured implementation program run with the vendor's legal engineers.

Lesson 1: Build a team, not a tool

LegalEdge isn't a software license. It's a small group with a mandate, a mix of skills, and real matters to work on. If your AI "program" is one enthusiastic lawyer and a browser tab, you don't have a program. You have a hobby.

  • Name an owner (even part-time) with authority to change workflows, not just buy tools.
  • Pair at least one lawyer with one person who thinks in process and data.
  • Give the team live matters, not a sandbox. Nearly 100 real matters teaches more than 100 demos.

Lesson 2: Decide what the AI may answer on its own

AT&T's announcement draws a clear line. For simple tasks, like frequently asked questions, an agent may respond directly. For complex matters, agents gather and synthesize information so a lawyer can give the advice. That's a triage rule, and you can write one this week.

  • Green: Policy lookups, process questions, template requests. Self-serve is fine.
  • Yellow: First drafts and research summaries. A lawyer reviews before anything leaves legal.
  • Red: Judgment calls, disputes, regulators, anything with the CEO's name on it. Humans lead, AI assists.

If you need a place to start, a legal intake and workflow tool is where triage rules actually live.

Lesson 3: Feed it your standards, not the internet's

BMW's first drafts start from approved internal standards and precedents. That's the unglamorous secret. A generative AI legal assistant is only as good as the playbooks, clause positions, and precedents you give it. If those live in five people's heads, the first project isn't "deploy AI." It's "write down what we actually believe."

Lesson 4: Measure quality, not deployment

AT&T's chief compliance officer, Bill Ryan, put it plainly in the announcement: "Success will not be measured by how much technology we deploy." Licenses bought and prompts run are vanity metrics. Better ones came out of a September roundtable of about 50 European in-house leaders, reported by Legal IT Insider on October 7. Delegates described tracking:

  • Time to first response
  • Time spent with external counsel
  • Total time to matter resolution

Those are numbers a CFO understands. If you're tracking outside counsel time, legal spend analytics will do the heavy lifting.

Lesson 5: Don't spend all your speed

One delegate at that same roundtable described a deliberate choice. AI could compress a week of work into an hour, but the team aims for a day and uses the rest for proper review. The business wants an answer in a day. It doesn't need one in eleven minutes with a hallucinated citation. Banking some of your speed as quality control is a feature, not a failure.

Lesson 6: Act like a design partner with every vendor

You probably won't get a joint announcement with an AI lab. You can still behave like a design partner. Bring real workflows to pilots. Agree on how you'll evaluate output before you start. Ask what's on the roadmap and who decides. Vendors listen hardest to customers who show up with use cases and data.

AT&T's general counsel, David McAtee, summed up the whole approach in four words: "Legal judgment remains human." Build everything else around that sentence.

Your Monday move

Pick one high-volume, low-risk request type, like NDAs, policy questions, or contract lookups. Write the triage rule, gather the approved standards, choose two metrics, and run it for 60 days. That's your LegalEdge. No press release required.

Aggregator X Webflow Template - Icon
CorporateLegal.tech

The CorporateLegal.tech editorial team covers the trends, tools and hard-won lessons shaping modern corporate legal departments.