AI for Small Firms



By Erich Dylus

Solo attorney (MD, DC) and programmer. Founder of:



Co-founder of a one-year-old girl who's made efficient work essential





What are we doing here?



We’re going to learn:


When used properly, AI has low downside and high upside as a time‑saver whose work you review

Why is everyone obsessed with AI suddenly?


Generating new data means it can create digital anything (text, code, images, audio, video, and paired with 3D printing... physical objects)



How was it built?

Huge amount of data → data cleaned → “trained” on patterns → extrapolate connections → optimize and finalize the model


Terminology

This all sounds too technical for me…



If you’ve used Google to search, you can use generative AI tools like ChatGPT




A genius intern is not going to replace you for years and can greatly reduce your workload in the meantime.



How can I use it, and how much does it cost?




If you’re a small or solo outfit like me, you’re not going to spend five or six figures on bespoke, in‑house models.

Fancy “Legal AI” services have little (if any) advantage without substantial customizations and maintenance (in-house data training, custom agents, periodic on-site updates).


Instead, I spend 1.5 burritos per month to subscribe to the most powerful knowledge generators in history.

All major LLMs provide powerful free versions with different specialties, and improve rapidly.



The magic is in your prompting abilities



Specialities change rapidly, but some services are purpose-built, like web search (Perplexity) and coding (Cursor, Windsurf)

Maintaining data privacy with all AI models, especially the free versions, is essential… more on this later.



How to Prompt



Role + Task + Instructions + Context



“You are an expert ___. Write a ___ with ___ and without ___. Review the attached file for context and ask any clarifying questions first.”



Invest the most time in your initial prompt to avoid backtracking later.

Remember, AI’s output is “probabilistic” — you’ll get different outputs for the same input (restaurant, not a vending machine).

Don’t be afraid to ask the exact same thing multiple times or to multiple models—or ask them to critique each other!

Context can also include your own writing, so the output is tailored to your style.



Bonus: XML Tagging

You can use XML tags to guide the model and more effectively structure the output. For example:


<role> You are an expert _______. </role>

<task> Create ________. </task>

<instructions>
1. Read the attached context file and this prompt in their entirety, and ask any clarifying questions before responding.
2. ______
3. Ensure any research you perform is cited.
</instructions>

<context> Attached is _______. </context>

Prompting Examples



Text analysis:




Form generation:





Negotiations, strategy, issue spotting


Less emphasis on the instructions, more on the role, context, and task.


This allows the model to be more creative and flexible, making it more helpful (especially as a mocked counterparty):
overly precise instructions can make it predictable and similar to your own thinking.


It may feel ridiculous, but try other human aspects that would affect the way you respond to a situation




Prompting Notes





Note the clarification requests.

Example categories can be overlapped (issue-spot areas of legal liability, then propose ___ ... analyze this, then generate a form ___).


If you want to mimic or follow the style or voice in your prompt (such as attaching one of your own writings), tell the model!


Don't stress the formatting— models usually use markdown that doesn't always translate well to your word processor.


Substantive but subjective prompts like these are even more effective with “reasoning” or "thinking" models.



What about Privacy / Confidentiality / Privilege?





Always assume the worst.

These are big tech companies with unprecedented compute power, and there’s a reason they offer free versions:


They use your data to train their models.


Even if they have a good data privacy policy, legislation, regulatory guidance, and judicial orders can force their hand:





These models don’t just collect and repeat data—they can extrapolate more information and even codify falsehoods.

Once models have absorbed and trained on data, assume it's impossible to “convince” it to forget or overwrite – this is true even for solely in-house AI models.

Conflicts-related ethical walls, or a terminated client relationship or GDPR compliance-related deletion request are problems.

So what’s the solution?

CamoText Demo



You need to ensure AI never sees private, confidential, or sensitive data.



Here’s a tool I built to more efficiently protect sensitive text





What Should I Tell my Clients?



Tell them the truth: you're using AI to be more efficient, without compromising your judgment, ethics, or privacy obligations.

If they’re using AI services, remind them of the severe hallucination and privacy issues.




... this is an actual response I received from ChatGPT in 2024.



If they're using them for legal-ish work, suggest using for general purposes, like honing questions and requests for you.

Win‑win: efficient intake, and clients appreciate attempts to reduce the bill (and empower them to learn about the matter)


Share the best practices with them!



Best Practices



Parting Words



An unbelievable amount of intellectual power, with near-instant response time, and reasoning + research capabilities, for comically cheap subscriptions.



You’re not getting replaced by AI anytime soon—you’re getting superpowered.

More actual legal work rather than repetitive keyboard mashing.



However, you might get outpaced by others using this technology if you don’t!

Low downside (takes seconds and you're reviewing the output anyway) and extremely high upside.



Also, it’s fun.

Talk to it like an interactive academic course or a podcast, learn a new skill or language with an infinitely patient and personalized assistant, have it create an outline for a presentation...



Links and Contact Info




Websites, Profiles, and Emails:

Other AI services mentioned: ChatGPTPerplexityClaudeGeminiGrok