Jesus Vargas

First time I hear about someone using Google models..

Preparing a few materials before you meet a development partner helps keep your project scope steady, because more decisions get made before the build starts. Discovery is the planning phase before the build, and it moves faster when the team can see how your work runs today. A few materials make that easy to show. First, record short screen videos of your main tasks. Second, write the steps in order, and add the exceptions you handle. Third, collect sample documents and screens from your current system. You can start without any of them, and we build the materials with you when you don't have them yet, right? Discovery at @LowCodeAgency covers process mapping, wireframes, project documents, and an estimate from our tech lead. DM me "process" and bring whatever you have, even if it's rough.

Healthcare builds start with scope. Confirm where HIPAA applies, review third-party tools that handle sensitive data, set audit trail retention to match policy, and keep clinical decisions with providers while AI supports documentation. Karoline from @LowCodeAgency walks through the checklist in the video.

Do you need an AI agent? The answer depends on how your work behaves. If your steps are the same every time, build an automation, because it costs less and runs more reliably. If people ask questions and need answers from your documents, a chatbot fits. If the work changes from case to case and needs research, tools, and checking, an agent fits. Sometimes an off-the-shelf tool covers all of it, and I'll tell you so on the call. When something super custom makes sense, we start with Discovery. It's 7,500 bucks and takes two to three weeks. We map your processes with your team, and our tech lead estimates the build. At the end, you can see what we're building, how long it takes, and what it costs. Send me a message with the process you're thinking about, and I'll help you figure out which one fits.

Here is what Rising Ground, a New York nonprofit we work with, has after we built their issuance app: 1. One dashboard where every item has an owner, a live status, and a history. 2. Requests that route automatically to the right team. 3. A digitally signed letter of responsibility that closes every handoff. 4. A retrieval sequence at every exit, confirmed by HR, where each item gets resolved one by one. The flow came from sitting down with their team, figuring out how people really work, and building around that. DM me and I will walk you through it.

Can you build it yourself in Claude? Maybe. Try it for a week. AI tools agree with you. Ask if your idea works, it says yes. Ask why it won't, it says yes. So a prototype feels 90% done when it's closer to 40%. The test: a month later, is anyone still using it?

Before we build anything, we identify the champions in each department. They volunteer first. They test things between calls. They become who their colleagues turn to when something needs explaining. We invest in them throughout the build. Dedicated Slack channel, office hours, consistent check-ins. After the transition, they own the feedback loop and keep everything running. We find them in the audit. We invest in them during the build. Six months later, they are the reason adoption holds.

Recently, I asked AI to line up every blog post we have published with the data from Google Analytics and Google Search Console. Each platform gave us hundreds of rows, sometimes thousands, and in Excel, that comparison means building a pivot table by hand. AI completed the whole comparison for us. Two years ago the focus was on automating individual steps with fixed rules. Today you can hand an agent a bigger process and give it the authority to make certain decisions along the way, and customer support is now in that group. The skill is figuring out which decisions an agent can own and which ones stay with your people, and a lot of that comes from mapping how the work runs today. Which task on your weekly list would you hand to an agent first?

AI coding tools shorten parts of a software rewrite, which is the rebuilding of an existing application on current technology. The time saved depends on the application. A simple application with few connections to other systems moves faster than one with a lot of connections, so each estimate needs to look at your software in particular. If a rewrite was quoted to you a few years ago, ask for a new estimate. First, ask which parts use AI coding tools. Second, ask which parts still need human design and testing. Third, ask how both answers change the timeline. A clear answer shows you what you're paying for, right? Sometimes your current system still fits and a rewrite can wait, and I'll tell you that. Bring an earlier quote to a call with me and the @LowCodeAgency team, and we'll go through it together. DM me "rewrite" to set it up.

Rising Ground. Fleet operations across New York. All of it tracked in Excel before we built their system. Jordan's words after launch: "What used to be spreadsheets now feels cleaner, more accountable, and easier to manage." One platform. Third year of continuous development.

The assumption most healthcare clients bring: compliance will slow us down. The pattern when compliance enters at the refinement stage: the decisions that would otherwise become blockers get made before they block anything. Which vendors will sign a BAA: decided before the architecture depends on one that will not. Which data fields are PHI: decided before the schema is designed. Which audit log requirements apply: decided before the storage architecture is finalized. Compliance built in from the design stage removes the stops.

Every email our AI agent writes gets approved by a person before it goes out. Here's how the pieces fit together. The language model reads the situation and chooses the next step. The tools do the actions, such as searching job posts, looking up the hiring manager, and saving the draft. A person approves each email at the end. AI output varies from case to case, so a review step makes sense for any message that reaches a customer. Every company starts with a bunch of open questions about agents, so we start with Discovery. We map the process and figure out which steps the agent runs and which steps a person approves. In the first couple of months of a build, the time from your team is about an hour a week. Send me a message if you have a process where you'd like an agent to do the research and a person to make the final call.

Offboarding is where most equipment needs to find its way home. The manager thinks about workload, HR about paperwork, and the laptop waits for its turn. In the workflow we built for Rising Ground, a New York nonprofit we work with, everything moves once HR confirms the exit. Picture someone on a three month medical leave: with HR as the single trigger, their email and badge access stay in place. Then the app lists every item assigned to that person. An admin marks each one as returned or documents the reason it is still pending. Notifications continue until every line has an answer. DM me and I will walk you through it.

Should AI change your records, or only read them? Start read only. Add write access one role at a time. Log every change and why. Give the AI the same limits as the person it works for. Changing a deal amount is not like adding a note.

Every client gets the same advice from me before we start: expect iteration. The first version is just the starting point. Each pass adds context and the output gets sharper. By the third or fourth round, the workflow runs clean and the team trusts what it produces. We build and train at the same time. By launch, the team has already been through multiple rounds. Every time they use it, it gets better. That compounds over time.

AI agents need the decisions your best people make written down. On day one, a new team member learns what to do at each step and how to decide when the next move needs some thought. An agent needs the same information. Your best people carry it, and they act quickly when the answer is clear and take their time when it needs some figuring out. Writing down both kinds of decisions gives you a guide for a new hire and for an agent. A short conversation with your best people is a good place to start. At the end, you can see where an agent decides on its own and where your team keeps the final word. Which decision would you document first? We're happy to walk through it at @LowCodeAgency.

Reviewing the manual steps in your daily work is a quick way to give your team time back. A lot of teams have a few places where people re-enter data that already exists in digital form, e.g., typing details from a PDF into a form. Each step made sense when someone set it up, and software can now handle some of them. First, list every place where a person types data that already exists in digital form. Second, mark which entries a program can read automatically, e.g., fields in a PDF or an email. Third, have a person check the results the program is unsure about. AI reading accuracy varies by document type, so a person still reviews the results, right? At @LowCodeAgency, we map your manual steps with your team during discovery. Sometimes a prebuilt tool or a simple form covers the need, and I'll tell you that on the call. I'm happy to talk it through when you DM me "manual."

Nick here is the Jim Cramer of X

The FDA is moving toward continuous monitoring requirements for medical devices with integrated models. Getting the model right at launch is the beginning of the requirement. Building the infrastructure to know whether it is still right six months later is the ongoing one. Most teams are building the recommendation. The audit trail for the recommendation is not on the roadmap. The FDA is telling you the audit trail is the product.

Here's what a custom AI agent for sales research does, step by step. We built it for an employer of record. The goal is to find US companies about to hire in Mexico and write the first email. First, it searches job posts for US companies hiring in Mexico. Second, it removes companies that already have a Mexico office. Third, it finds the hiring manager. Fourth, it picks the closest case study. Fifth, it writes the draft email. The full version runs about 90 steps, and the agent chooses the order based on what it finds. A person reviews each draft before it goes out. Researching one prospect by hand takes a bunch of tabs and a lot of checking, and the agent does it for the whole list. What goal would you hand to an agent in your company?