From Prompts to Playbooks in Commercial Real Estate

Every week, property teams receive the same breathless pitch: AI will transform your practice. A year ago, the practical advice that followed was almost always about prompting — learn to write a good prompt, and you're set. That advice is now out of date. Prompting is still a real skill. But it is no longer the unit of durable work. The unit is the skill itself — built once, tested, and run the same way every time, by anyone on the team, without anyone needing to remember how it was originally written.

This is not about the hype. It is about building AI workflows that hold up under pressure — ones a paralegal can run on Monday morning and a partner can trust on Friday afternoon, without either of them retyping instructions into a chat window. The discipline applies equally whether you are an institutional landlord managing a shopping centre portfolio or a retail brand operating across multiple markets.

Why a Good Prompt Isn't Enough Anymore

A well-crafted prompt is still a specification. It tells an AI who it is, what the task is, what the output should look like, and what the constraints are:

Role — "You are a senior commercial real estate attorney advising a US institutional landlord."

Context — "This is a triple-net lease on an inline retail unit in a regional shopping center. The tenant is a national apparel brand."

Task — "Summarize the key repair obligations, rent escalation mechanism, and assignment provisions."

Format — "Return bullet points under three headings. Flag anything that departs from ICSC standard market practice."

That's a good prompt. The problem is what happens next. Somebody has to retype it, correctly, every time — and remember the constraint that stops the AI from citing a case it isn't confident is real, and get the tone right for this particular reader, and hope nothing was left out on a Friday afternoon with a closing deadline. A prompt that lives in one person's memory or one person's saved notes isn't infrastructure. It's a habit, and habits don't survive turnover, busy weeks, or a new hire who's never seen it.

Building Skills

A skill is what a prompt becomes once it stops depending on any one person typing it correctly. It's the same specification — role, context, task, format, constraints — built once, tested against real documents, and packaged so it runs identically whether the partner runs it or the paralegal does.

What a Good Skill Actually Does

Not everything marketed as "AI-powered" is a real skill. Here's what separates one from a general-purpose assistant improvising in a chat window:

It carries its own expertise. The role and context aren't retyped each time — they're built into the skill itself. Nobody has to remember to say "you are a commercial real estate attorney" before every lease review.

It produces the same structure every time. A risk summary should look like a risk summary whether it's run on Monday or six months from now, by whoever happens to be at the keyboard.

It's grounded, not just fluent. Every extracted term should trace back to a clause and a source, so a reader can verify it in seconds instead of trusting it on faith.

It knows what it doesn't know. A good skill flags ambiguous language and low-confidence extractions for human review, instead of quietly filling gaps with a plausible guess.

It has explicit guardrails. "This is for internal research only. Where you are uncertain, say so. Do not cite cases unless you are confident they are real." That kind of instruction shouldn't depend on someone remembering to type it — it should be permanently built into the skill.

It's owned and maintained. Someone is accountable for it, updates it as edge cases surface, and retires it when it stops earning its keep. A skill nobody owns degrades quietly until someone notices the output is wrong.

Building Workflows

A single skill does one job well. A workflow chains several skills together, with the output of one becoming the input to the next — the same discipline as decomposing a task, just built as infrastructure instead of re-derived by hand each time.

A lease review workflow might run as five linked skills:

An extraction skill pulls commercial terms — parties, premises, term, base rent, percentage rent, co-tenancy and kick-out clauses.

An obligations skill separates landlord and tenant responsibilities.

A deviation skill flags anything that departs from house standard or local market norms.

A risk-summary skill drafts a risk-rated memo suitable for an investment committee.

A drafting skill produces the cover email to the client, in the right tone for that reader.

By the time the workflow reaches the fifth step, every prior step has already been validated. Junior team members can own individual steps without needing to understand the whole chain. Quality checks happen at each stage, not just at the end — which means errors get caught where they happen, not three steps downstream.

The same pattern extends well past lease review. Reciprocal easement agreements, operating covenants, estoppel certificates, SNDAs, construction side letters, and guaranty agreements all benefit from the same treatment — especially letter agreements, which are notorious for being filed and forgotten. A buried rent abatement right or co-tenancy cure period, surfaced automatically before a renewal negotiation, can be worth real money. ESG and sustainability reporting is another natural fit: a workflow that extracts green lease provisions across a portfolio, flags leases with no utility-data-sharing framework, and drafts investor-ready ESG narratives turns a manual quarterly scramble into something that runs on its own schedule. Precedent comparison — pulling your last five leases for a similar asset class and producing a comparison table across fifteen key provisions — becomes a workflow your team runs in minutes instead of a task someone dreads starting.

Automating Workflows

Chaining skills by hand is a real improvement over ad hoc prompting. But the workflows that actually compound value are the ones that don't wait for someone to open a chat window at all.

An amendment gets executed — the abstraction workflow updates the rent roll and critical dates calendar automatically. A new LOI arrives — the comparison workflow runs against the draft lease without anyone pasting documents in manually. A portfolio acquisition closes — the full due-diligence pipeline kicks off across every lease in the data room on its own, flagging what needs a human look and clearing what doesn't.

This is the real difference between a playbook and a habit. A playbook that lives in a shared document still depends on someone remembering to open it. A workflow wired to the events that already trigger the work — an amendment, a new deal stage, an acquisition close — runs whether or not anyone remembers to ask. The teams that get there aren't doing more prompting. They're doing less of it, because the trigger does the work a person used to have to initiate by hand.

Where Bryckel Fits In

Here's the part most teams underestimate: building and maintaining skills like this — testing them against real documents, versioning them, wiring them to the events that should trigger them, fixing them when a new lease format breaks an assumption — is a full-time discipline. It is not a broker's day job, a lease administrator's day job, an asset manager's day job, or a legal team's day job. It's Bryckel's.

For brokers, that means deal-killer checks and LOI-to-lease comparisons that turn hours of review into minutes, plus market briefs that update on a schedule instead of a scramble.

For lease administrators, it means abstraction, critical dates tracking, and rent roll maintenance that update the moment an amendment is executed — not weeks later during the next portfolio audit.

For asset managers, it means portfolio performance dashboards, CAM reconciliation review, and risk-rated summaries ready for the investment committee without a manual pull-together every quarter.

For legal teams, it means redlines run against your own playbook, NDA triage, precedent comparison, and research workflows with grounded citations and the guardrails already built in — not retyped into every prompt.

In every case, the skill runs inside the AI environment your team already has — Claude, Microsoft Copilot, or ChatGPT and Codex — on the model your organization already licenses. Your model. Your workspace. Our expertise.

The Shift That Matters

You don't need to automate everything at once. Start with one skill. Test it against real documents. Wire it to the trigger that should run it automatically. Then build the next one.

If your team would rather have that built and maintained than build it yourselves, that's exactly what we do. Reach out and we'll show you what a skill built for your playbook looks like, running inside the tools you already have open.