← All webinarsInsights home
Webinar recap · On demand

How enterprise legal teams put agents into drafting workflows

Missed it, or want to go back over it? Paul Lacey's live session on handing complex drafting to agents: what the business user sees, why reading the deal and assembling the document are two different problems, how a lawyer supervises the result, and a first look at turning an annotated Word template into a working agent. The full recording is here.

Paul Lacey VP of Product, Flank · with Jake Jones
Format Live demo + Q&A
Recorded Wed 21 October 2026
Read time ~6 minutes
The short version Watch The business user Two problems Supervision Templates Takeaways
In one minute

The short version

None of the individual decisions in a complex contract is hard. Every deal has a term somewhere, and whether a clause applies to a given brand is a simple call. What makes drafting hard is holding hundreds of those decisions at once, across a hundred pages and a deal whose facts are scattered. That is why it still lands on senior lawyers.

The session was a live demo of how Flank's customers take that work off them. Not NDAs or order forms: franchise agreements and other core contracts with layered conditional logic and messy deal data. The approach splits the job in two. Agents read the deal, chase what is missing and surface conflicts. Proven template automation assembles the document, so the same facts return the same contract every time. A lawyer signs off a short list of flagged inputs rather than re-reading the whole instrument.

Read
This recap
The problem, the architecture, the supervision view and the template tooling, in plain text.
Start reading ↓
Watch
The recording
The full session, including every live demo and Jake Jones's framing of why drafting is still unsolved.
Watch the session ↓
Try it
On your own template
Bring the agreement nobody enjoys owning and see what an agent makes of it.
Book a demo →
Watch

The full session

Jake's introduction, Paul's live demos and the closing Q&A. Everything shown runs on live systems, so the occasional pause is real. If you would rather skim, the written recap below covers the substance.

The frame

Why drafting still isn't solved

Drafting tools have been around for over a decade. Jake opened by asking why senior lawyers are still assembling the same agreement for the fortieth time with a different set of facts. His answer: three things the templating tools never touched.

01
The work around the work
Reading the heads of terms, chasing the business for the three facts nobody included, reconciling two values for the same input, routing approvals, checking existing deals, getting it signed. A template does none of that, so a person does, and the tool becomes one more step that runs through a lawyer.
02
Setup and maintenance
Turning implicit drafting knowledge into programmed logic is hard upfront, and what is left behind is a puzzle. Teams end up with a dedicated automation team, or with templates they can no longer update.
03
Not where the business works
Portals and request forms go unused. The request still arrives by email, at an odd hour, missing the context needed to draft it.

Solve those three, Jake argued, and the picture changes: an agent collects the request by email, chases what is missing, checks it against Salesforce or the CLM, drafts the whole document set, checks it against house positions, gets approvals, sends it for signature and files it. The lawyer supervises the system and approves the points that matter.

Demo 1 · The business user

A franchise agreement from one email

Paul started where the workflow has to start: with the person who needs the contract. If the business does not adopt it, nothing else matters.

Request by email, to the inbox they already use. Flank plugs into existing shared mailboxes and routes relevant requests to the right agent. There is no new address or portal to learn. Paul sent a request with an unstructured heads of terms attached.

The agent comes back with what is missing. It replied with a pre-populated form showing every deal fact it had found and where each one came from: the email, the heads of terms, or Salesforce. Governing law was missing. Paul picked New York, and the full agreement came back in under a minute. Most of the agent's effort goes into understanding the shape of the deal before it drafts, so nothing is ambiguous when assembly starts.

Drafting from inside Salesforce, and back again. For one customer, the deal team triggers the draft straight from the Salesforce opportunity with a pre-built email that carries the deal data. Once negotiation ends, they email the signed agreement back and the agent updates Salesforce to match what was actually agreed, closing the gap that opens during redlining.

What changes for the business

Wait time drops from days, sometimes weeks, to a couple of minutes. Hours, where a lawyer reviews before it goes out. That is the number the deal team notices.

The architecture

Two problems, two technologies

Paul was clear about what this is not: pasting a template and some deal notes into a chatbot. That output looks convincing until you check it. Flank splits drafting into two problems and gives each to the technology that is good at it.

01
Reading the deal
Heads of terms, email threads, CRM records, previous agreements in SharePoint, even dozens of org charts to work out an ownership structure. This is the messy part, and it is where AI agents are strong: they go back to the user, ask for what is missing and surface inconsistencies.
02
Assembling the document
Variables and conditional logic in a template. This was solved years ago, and it is deterministic: if the inputs are right, the output is right, ten times out of ten.
Where the AI earns its place

The heads of terms says 100 rooms. Salesforce says 105. Which is it? Old deterministic systems cannot handle that. An agent can raise it with the user and ask, rather than quietly picking one and burying the choice in a clause.

The signs that a team has a strong candidate for this: lawyers doing the assembly, and in some cases a second team set up purely to catch the first team's mistakes.

Demo 2 · The lawyer

Reviewing the inputs, not the whole instrument

These are business-critical contracts, so an agent cannot send them out on its own. A lawyer has to check the work and take responsibility for it. Supervision is designed around that.

Every request lands in one inbox in the Flank web app, connected to the channels legal already uses, so you can see what is with an agent, what is with a person and what is waiting for review. Opening a drafted agreement shows the contract on the left and a review panel on the right with two tabs.

01
Inputs
Every variable in the contract, with its value and its source. Flagged items sit at the top with the reason they were flagged: either a field you always want to check, or one where the agent's confidence was lower. Confirm or override, and jump to where it sits in the document.
02
Variable clauses
Which version of each conditional clause was chosen, and what it depends on. A hotel development clause, for example, turns on property status. If the choice is wrong, you click back to the input that drove it.

Once approved, the reply to the business user is drafted and sent. Some customers skip the panel entirely: they email the agent, get the draft back with track changes against the template, and accept or edit inline. The review step stays the same either way.

Demo 3 · Configuration

From an annotated Word template to a working agent

The part Paul was most excited to show, and the one that answers the maintenance problem. It is early stage, and it was demoed live.

When new customers send a template, it usually arrives as a Word document with highlighted variables and comments explaining the logic: "this clause is for all brands except one". Turning that into template syntax used to be slow, manual and error-prone, for customers and for Flank's own team.

A Word add-in now reads the annotated template and works out the variable fields, the rules and the conditional clauses. For each field it captures what the agent needs to find it: type, whether it is required, where to look (the heads of terms first, then the covering email) and what to avoid confusing it with. It flags anything it is unsure about, applies the logic to the document, and exports the instructions and the structured template straight into the agent's configuration. A preview panel lets you feed in sample deal data and generate a test document before anything goes live.

Why it matters

Getting from zero to one, and staying at one. Setup that took hours of careful manual work now takes minutes, and iteration loops shrink from days to hours. Teams can maintain their own templates instead of depending on a vendor for every change.

If you remember five things

The takeaways

  1. Complexity is volume, not judgement. No single decision in a complex contract is hard. Holding all of them at once is, and that is work you can hand to a system.
  2. The work around the work is where the time goes. Chasing facts, reconciling sources, routing approvals and filing. A template alone automates none of it.
  3. Split the problem. Agents read the messy deal; deterministic templates assemble the document. Either one alone falls short.
  4. Conflicts should be surfaced, never guessed. When two sources disagree, the agent asks. The lawyer reviews a short list of flagged inputs, each with its source.
  5. Meet the business in its inbox. Email and Salesforce, not a new portal. Adoption decides whether any of this pays off.
Three questions to take to your GC

01. How many documents does a single deal produce?

02. Where does the deal data live: one system, several systems, documents, email threads?

03. Who assembles it?

If the answer to the last one is a qualified lawyer, you are paying judgement prices for repetitive assembly work.

See it on your contracts

Bring the template nobody enjoys owning. We'll show you what an agent does with it, from the first email to a document a lawyer can sign off.

Book a demo

Subscribe

The Intake

Weekly briefings on what's actually changing in legal AI: the market shifts, the regulatory moves, and the structural questions that matter for enterprise legal teams. We'll also send details of the next webinar.

Subscribe on Substack
Flank

Outsource legal work to supervised agents

Enterprise legal teams use Flank to handle high-volume contracting end to end: NDAs, MSA redlines, procurement, and triage. Agents that know your templates, your terms, and your escalation rules.

Learn more at flank.ai