Just Say It. Chapter by chapter.
Our keynote in 13 short chapters. Watch only the part you need, or press Play all for the whole show (about 39 minutes).
Read the transcript of this chapter
Good morning everyone and welcome. My name is Shlomo Friman, IN-COM Data Systems. And behind the scenes there is Ami Sadeh, CTO, and his entire amazing development team. Hey guys, thank you. Today is the day we've been working towards for a very long time. For many years, you've trusted SMART TS XL to help you understand your code. Millions of lines, some of you billions of lines. Mainframe, distributed, COBOL, Java, databases, all of it. You search, you explore, you find, you analyze, flowcharts, complexity. Many of you have become true experts of what we call power users and thank you for that. Today, something changes and it's big. What if you don't have to search at all? What if you can just say it? Imagine this, if you will. Take out your phone, open Claude and say, Where's tax code updated across every platform and every system in the company? Boom, you get it. What if you could say, Give me a list of every COBOL program that opens files without checking file status? Or, if we were to retire these three datasets, what's the ripple effect? What's going to break? You say it, SMART TS XL does the work for you. It's that simple. It's not just chatbots guessing here. It's the AI you're already using and trust, like ChatGPT, Claude, Copilot, Gemini, working hand in hand with SMART TS XL and its MCP server. Every answer traces back to the exact line of code, evidence. And it learns, correct it once, and from then on, the whole company gets the right answer. And because the AI only reads what it needs, your AI bill goes down dramatically. 20, 40 percent or more. Now, I know not all of you have seen this or have this yet. That's okay. It's coming very soon. Today is about showing you what's possible, what's just around the corner, one, two weeks away. So we're going to do something a little different. Instead of starting with what you could do today and building up with a big wow, we're going to start with the wow and then show you how we got there. One caveat, this lecture is 30 minutes from start to end. Minute 31, this is going to be a reveal of the latest technology we are now working on and you will be the first to see it. All right. Let me show you now what happened this morning as I was driving to work and how I could make Just Say It work for me. Let me show you.
This example, we're going to talk about the power of connecting your favorite LLM to the MCP connector of SMART TS XL. You can see that the MCP of SMART TS XL grants access to the LLMs to perform all the major skills, tasks, features and functions of SMART TS XL. So basically, you have the LLM with all the knowledge of how to operate SMART TS XL to perform any task we want. Now, let's look at the task. Let's look at the task at hand. While you're driving, you open your phone, chose, in this case, Claude, and said, expand the tax code field, WS tax code, from two to three bytes. Find every program, copybook, JCL job and file it touches. Tell me what has to change, how complex each program is, and how many hours it will take. And give me a phased project plan with the risks. Oh, and I want that color coded report in my inbox by the time I get to the office, please. So you can see that it started working immediately. And as I was driving safely, not speeding to the office, take my time, it finished and generated a PDF report. This PDF report I then asked to email to my emails. And let's go have a look at what I was emailed. I'm going to my Outlook, open the email, and look at this file he created. You can see that the file is color coded, easy to read. It's the expansion to three bytes. It tells me how many hours, how many days, how many complexity programs, the fields it found, the affected fields it's moved or moving to is here. Also found all the affected programs, and it tells you if they're medium, high, medium, or low complexity analysis. And has a project plan of exactly start, design, code, test, deployment, and a summary of risks and open items. So this is definitely a game changer when a user can simply speak his heart to an LLM and specify exactly what he wants. And then the LLM together with the SMART TS XL MCP can use all the tools and features and functions available to SMART TS XL to perform even monumental tasks. It might take a minute, it might take two minutes, but we're talking about saving weeks and months of work on every ask. I hope this helps. And anyone interested in more information, feel free to reach out. Back in the day, developers had a saying, what you see is what you get. Well, with SMART TS XL, we coined a new one, and it's become our little thing. What you say is what you get. So how did that magic happen?
How is it that I can just speak into my phone or my computer or simply type what I want and SMART TS XL does all the work for me? Let's dive in. Everybody knows what an LLM is. ChatGPT, Claude, Gemini, Copilot, a super smart AI. And everybody knows SMART TS XL, your entire code landscape indexed and searchable in seconds. But the magic sauce is in the middle, the IN-COM SMART TS XL MCP server. MCP stands for Model Context Protocol. Big words. So what does it really mean? Between you and me, it's a middleman, a middleman who knows everything there is to know about SMART TS XL. And I mean everything. Even better than Ami and me, every feature, every function, every tool, search, synonyms, cross-reference, reports, flowcharts, field tracking, it's all built in. How? We at IN-COM gave it context files. Everything to know about SMART TS XL, and everything to know about you, the client, what you've loaded, your data types, production and test, mainframe and distributed. It knows all of it. Now, you give the LLM a big ask. If I change this field, what's the ripple effect? Show me every place that has to change and give me the whole project plan. The LLM talks to the MCP and together they work out a plan. To get you that answer, we need five searches, 18 cross-references and a complexity check. Then the MCP goes into SMART TS XL through our APIs and brings back exactly what the LLM needs. Millions, even billions of lines of code. And here's the part that saves you money. Normally, an LLM has to work really hard to dig up information and that burns a lot of tokens. In our case, the MCP does the running, the legwork, the heavy lifting, and it delivers everything to the LLM on a golden platter, saving you millions and millions of tokens. That, in a nutshell, is the MCP. Now that you know how it works, let's push it.
Three questions. Easy, harder, and really hard. Watch. First, an easy one. I type, find every place we calculate tax or interest. Claude asks the MCP and the MCP runs one search in SMART TS XL. Look at that. Almost 4,000 production programs searched in three hundredths of a second. 113 programs calculate tax. Over 2,000 hits. And interest? Almost nothing. Just one program. A real answer in seconds. Now a harder one. Trace the tax code field. Who uses it, what it touches, and how it flows across the subsystems. The MCP finds the field, runs a cross-reference two levels deep, and checks the jobs that run it. Seven programs share this field. One big hub program. Over 10,000 lines with almost 600 go-tos. And they're all tied together through shared copybooks and procs, not program calls. That's something you'd never see just by reading the code. And now the really hard one. Show me every link between the mainframe and the distributed systems for payment processing. With the risks. 16,000 documents searched. Then the MCP follows the trail. A JCL job runs a COBOL program. An MQ bridge reads the queue. A Java service picks it up, calls a .net worker, and writes to two databases. SQL Server and Oracle. Every hop found in the source. And the risks? The queue names don't match between the COBOL side and the Java side. The job names disagree. And the MQ read isn't protected, so a message could get lost. That's the kind of thing that takes a team weeks to find. Easy, harder, really hard. Same chat box. Just say it. Easy, medium, and hard were the warm-up. Now let's do the real thing. A real change request. On real, working production code that we sanitized from start to finish. Here's the requirement from my manager. The customer account type is two characters today. Like ST and PA. It needs to become three. Find every occurrence and every spelling. Trace every program, call, JCL job, data set, and sort it touches. Draft the code changes. And prove how many tokens SMART TS XL saves while doing it. I paste that request into Claude. Exactly as written. Claude is connected to the SMART TS XL MCP server. You can see it right here, under connectors. I press enter and away it goes. It loads the SMART TS XL context, then searches 146,000 members in about 3 seconds. Less than half an hour later, we have three deliverables. First, the token report. With the SMART TS XL MCP, the whole job took about 340,000 tokens. Without it, the AI has to read the source code itself. Almost 2 million tokens. That's more than fits in a 1 million token context. 83% saved. On a single change. These are very impressive numbers. In dollars, one change looks small. From about $4 to about $50 saved, depending on the model. Now, multiply that by a thousand developers, a hundred tasks a day, and AI agents running around the clock on your mainframe. That's tens of millions of dollars a year or more. This is fantastic, and it's going to do wonders for our clients. Second, the impact report for management. One page tells you everything. 13 programs, 3 copybooks, 53 jobs, 2 VSAM files across 8 branches, and 9 sorts. The ripple effect map shows exactly how the data flows. Every program with its size and complexity. The rating? Medium. The effort? About 28 person days. And it caught what a text search would miss. A hard-coded key with no field name and 18 look-alike fields that must not be touched. And the cherry on top? The suggested changes. One file for every program. Copybook and JCL member. 115 changes in all. Red is the old line. Green is the new one. With the reason. The copybook. The program. The proc, where the record length goes from 63 to 64. And the sort card, where the key goes from 8 to 9. Nothing in the source was touched. And think about it. We could have had the entire programs, JCLs and procs written ready for production review. This is really promising. One prompt. Less than half an hour. A complete impact analysis, the code changes, and 83% fewer tokens. That's SMART TS XL with MCP Server.
Okay. That was a lot to chew on. Really. A lot. So, let's take a little breather and have some fun. I've got three short clips for you. While you watch, think about which one you relate to the most. At the end, I'll ask for a show of hands. Now, these little stick figures may not look like much, but the problems they represent are very, very real. I've worked in big organizations and I can tell you, these things happen every single day. Let's have a look. Hey Tom, quick one. The auditors want every program that touches social security numbers. By Monday. Monday? Linda, it's Friday. It's 4.55. Honey, cancel the lake. I'll be home in March. Tom, put down the sleeping bag. Watch this. Find every program that touches social security numbers. Show me the fields, the files, and the risk. Send it to Tom. Wait, you can just say it? Just say it. Same SMART TS XL. And it does all the work. It's done? It's 4.57. Go home, Tom. Lake, here I come. Welcome back to Cooking with Code. Today we're making a classic impact analysis. You'll need 4,000 programs, 12 analysts, 300 sticky notes, and let it simmer for six weeks. Six weeks? Mamma Mia! Let me show you a shortcut. Run a full impact analysis on account number. Every program, every file. Plate it nicely. And here's one I asked for earlier. You just said it? Just say it. SMART TS XL does all the cooking. Okay team, how long to estimate the database conversion? Well, first we inventory every program. Then we check every table. Then we cross-reference everything. I'd say three weeks. Just for the estimate. Dave, turn around. Is that the estimate? Hours, risks, complexity. I just typed what I wanted, and it did all the work. So what do we do with the other three weeks? Start the project, Dave. Stop clicking. Start asking.
So here's the question I always get. Why not just use ChatGPT or Claude or any AI straight on my code? Why do I need SMART TS XL and the MCP? Great question. Because the difference is huge. Let me give you six reasons. One, faster. On its own, an AI has to read your code file by file. SMART TS XL searches over 100,000 documents in a third of a second. Two, cheaper. Without us, the AI reads the whole repository and you pay for every token. With us, it reads only the answer. 20, 40, even over 60% savings. Three, more accurate. An AI on its own can guess. With SMART TS XL, every answer is traced to the exact lines of code. There's always evidence. Four, consistent. Ask an AI the same question twice and you can get two different answers. With us, same question, same evidence, same answer. Every time. Five, complete. An AI only sees what fits in its window. SMART TS XL sees every program, every field, every relationship, mainframe and distributed. And six, smarter every day. Because we have something nobody else has. A self-learning MCP. Patent pending. Here's what that means. In every company, the real knowledge lives in people's heads. Did you know that programs starting with AB are payments? That modules ending in M are math routines? That the nightly jobs follow a naming rule nobody ever wrote down? Today, when those people leave, that knowledge walks out the door. With the self-learning MCP, the first time someone says, that's wrong. Here's how it really works. The MCP learns it. And from then on, the whole company knows it. So, why not just ChatGPT on your code? Because an AI that reads your code is not the same as an AI that knows it. Let me show you. Every company runs on knowledge that lives only in people's heads. Which programs belong to which application? What prod really means here? Your AI assistant doesn't know any of it. So it makes the same mistake today, tomorrow and for everyone. Here's a real example. Dana asks her AI for the ATM programs that update the card balance. It returns 40 programs, but half of them aren't ATM at all. Dana knows the secret. Here, ATM programs start with the letters AB. Nobody ever wrote that down. Now, meet the self-learning MCP for SMART TS XL. It's patent pending and it's already being built. It turns every miss into a lesson, right inside the AI tools your people already use. Claude, ChatGPT, Copilot and more. Here's how it works. The AI notices something is off because Dana says, that's wrong. It fixes her answer right away. Nine true ATM programs. Then it asks one quick round of questions. Does AB always mean ATM? Any exceptions? Should everyone learn this or just you? It takes about a minute. Here's the clever part. Before saving anything, SMART TS XL checks the claim against the real code. AB shows up in almost 500 mainframe programs, jobs and copy books. A few Java files only look similar, so they're left out. It's not a rumor. It's a verified fact. The lesson is saved as a simple card and it works for Dana immediately. To share it company-wide, an expert reviewer gets one email, sees the evidence and approves with one click. Personal preferences like show my results as a table apply instantly, just for you. Two weeks later, Yossi, in a completely different AI tool, asks which ATM programs call the fraud check. He's never heard of the AB rule. He doesn't need to. The AI already knows, gets it right the first time and shows which lesson it used. One person taught it once. Everyone benefits. And it's built on trust. A lesson can never give anyone access they don't already have. Conflicting lessons are caught, never silently stacked. If a lesson goes out of date, anyone can challenge it and an expert decides. Every change is recorded. For the organization, this is huge. Expert knowledge stops walking out the door when people retire. New hires get expert answers from day one. Fewer wrong results, fewer retries, less wasted time, on mainframe and distributed code alike. And the more people use it, the smarter it gets for everyone. SMART TS XL with the self-learning MCP. Patent pending. Already in development. Your company's knowledge. Learned once. Shared with everyone.
Let's talk about money. Your money. But first, a little movie. Meet the AI Bill. Tom, why is our AI Bill bigger than the coffee budget and the rent? It's the AI. It reads COBOL, JCL, Java, Python, database scripts to answer one question, then forgets and reads them all again. So we pay it to read the same code. Every single time? Yes. Watch this. What breaks if I change the zip code fields? It didn't read everything. It already knew where to look. Same AI. Same questions. Smaller Bill. Stop paying your AI to read. Start asking. It's a funny little movie. But this is a serious topic, people. When your company starts using AI a lot, it gets expensive. Very expensive. We know clients that are spending hundreds of millions of dollars on it. Here's why. You pay for the reading, not for the answer. Give the AI a big job. A request that's 50 sentences long and on its own, it searches. And searches again. And again. It opens programs, reads files, forgets and reads them all over again. And every page it reads, you pay for. Now watch what happens with the SMART TS XL MCP. The AI starts talking to the MCP and the MCP says, Whoa, wait a second. Hold your horses. Save your tokens. I'll get it for you. Everything SMART TS XL knows is already there. Millions, even billions of lines of code. Every rule, every relationship, every cross-reference, every data flow. The complexity analysis. It's yours. You own it. And asking it doesn't cost you a single token. So the AI asks once and gets back just the answer. That saves you anywhere from 20% to well over 60%. I want to save you money. That's the bottom line. So what does that mean for a company your size?
Whether your company has tens of millions, hundreds of millions, or even billions of lines of code, a big part of your AI bill is the AI reading that code. Whether your code still lives on the mainframe or you've already moved it into GitHub or GitLab, your AI works the same way. Search. Open a file. Read it. And pay for every line it reads, again and again, every time it looks. SMART TS XL has already read, understood, and connected every line on every platform in every language. Through our MCP server, ChatGPT, Claude, Copilot, Gemini, or Grok simply asks SMART TS XL instead of reading. First, exact lines in seconds. Not whole files, just the lines that match with the lines before and after for context. In our tests, we searched 131,000 documents in a third of a second. Second, questions that no text search can ask. Complex Boolean logic. Words near each other. Synonyms. Whole blocks of code. Code only, not comments. Even fields you define yourself. One precise question. Instead of thousands of noisy matches, the AI has to read and throw away. Third, SMART TS XL already knows every relationship. Which job runs which program? Which program calls which? Which copybook? Which queue? Which table? Which trigger? Across mainframe, Java, .NET, Python, and databases. The AI doesn't rebuild the map by reading, it just asks. And complexity, impact, and history are already calculated, so every new AI session starts smart instead of starting over. The savings. Well over 60% on the mainframe. About 20% on other repositories. At 20%, spend $10 million on AI, and you save $2 million. Spend $100 million, save $20 million. Spend $1 billion, and you save $200 million. Your AI already reads your code. With SMART TS XL, it reads the answer, not the whole repository. See what you'd save. Book your live session at in-com.com/savings.
That envelope stays sealed a little longer, but you don't have to wait for everything. In the version you already have, AI is inside SMART TS XL today. Help chat, feedback, and ask AI. Let me show you. First, brand new: Help chat. You can chat with Claude directly from inside SMART TS XL. It helps you build your searches, and it teaches you the product as you go. New in SMART TS XL, the Support Expert chat. A product expert you can ask anything built right into help. To open it, click Help in the top menu. Then, under User, choose Support Expert Chat. It answers from the official SMART TS XL documentation and IN-COM's own support knowledge for the version you're running. You can even attach screenshots and log files to your question. Let's ask a real question. How do I find every place we move a single character into a field with tax in its name? We click Send, and it looks it up in the documentation. Here's the answer. A proximity search. Move, then two, then any field containing tax. It explains why two needs quotes and how to keep comments out of the hits. And it's honest. It tells you when it's combining documented features, so you know to test it. Click Help, choose Support Expert Chat, and just ask.
Second, feedback. Send us feedback from inside SMART TS XL. And the moment you do, you get an email. We got it! And right there, the reply from our AI expert support agent. No waiting. Here's something new in SMART TS XL. When something doesn't work the way you expect, you can tell us right from the screen you're on. And within minutes, you get an expert answer by email. Let's try it. We search our code for the word read, and nothing comes back. That's not what we expected. So we click the feedback button right in the top bar. We pick the area, search. We choose the sad face. And we tell IN-COM what went wrong in our own words. Send to IN-COM. A quick thank you, and we're back to work. Behind the scenes, a ticket opens in our service desk automatically. Our new AI support agent reads it, checks the SMART TS XL user guide, and our past support answers, and writes back. It's like hearing from our best support engineer. Right away in the part that's new. An answer from our AI support agent. It tells you how it understood your message, and asks you to correct it if that's wrong. So every piece of feedback does two things at once. It opens a real ticket that our support team follows up on, and it gives you an expert answer right away, so you keep working instead of waiting. Report it. Get an answer. Keep working. That's support built right into SMART TS XL.
And third, ask AI itself on a single line of code, on an entire program, on many programs at once. Pick your model, use our ready-made prompts, or write your own. Have a look. There's a library of ready-made prompts grouped by task. Understanding and documenting code, debugging and optimization, modernization, and flowcharting. Click the pencil to see exactly what a prompt says and edit it to fit your needs. Open Ask AI from the results, and it works on your code. Interactive runs on one program, and the answer comes right back on screen. It's the best way to test a prompt. Background runs the same prompt across many programs, or all of them. Now choose what the AI reads. All the lines of each program, or only the search hit lines, or the hit lines plus context. Five lines before and after, or as many as you need. Fewer lines mean sharper answers and lower costs. And with the Gemini image creator, Ask AI draws. Pick the business rule flowchart prompt, and you get a picture of how the business rules in a COBOL program work together. Or choose an expanded JCL job and get its whole flow. The job start, every step, every proc, every return code check, right to the job end. Download the image and use it anywhere. Ask AI in SMART TS XL. Any model. Your prompts. Your code. Available today in version 13.
You've seen the MCP. You've seen it learn from every correction. You've seen one sentence from a car become a whole project. That was the beginning. What comes next? More exciting. Bigger savings and cutting edge technology. You are the very first audience to see what I'm about to introduce, and I couldn't be more excited to share this moment with you. Introducing SMART KS, the smart knowledge system. At IN-COM, our commitment is to help you lead the way with AI, to see what's possible, put it to work, and shape what comes next. And we want to build that future with you. Today, you're getting more than a first look. You're here at the beginning of something we want to create together. Together, we will define how IT teams work in the years ahead. This is huge. Deeper learning. Greater self-improvement. Even bigger token savings. This new development will take the SMART KS system even further, dramatically reducing token usage to make your AI work faster, deliver better results, and cost substantially less. And there's more. Another patent is pending for this innovation. Let me show you what we've been building. Until now, your AI assistant reached SMART TS XL through our MCP. It asked the question, got back the exact lines, and did the rest of the thinking itself.
Now, there's something much bigger. Meet the SMART KS agent. KS stands for knowledge system. The agent can do everything the MCP can do, and more. But it doesn't just answer questions. It does the work. Inside it are four things. And right next to it, one more. One, every SMART TS XL tool. Search, cross-reference, field tracking, complexity. Every program, every relationship, across mainframe and distributed code. Two, certified skills. Each skill is a program, a capability, pre-built and tested by IN-COM. It does a specific task and gives the same result every time. The agent comes with a library of them. Some can be built for one client. And the library keeps growing. Three, a coding sandbox. When a job needs something new, the AI writes a small program in any language and runs it safely inside the agent. That program is saved. The next time it's needed, the agent reuses it instead of writing it again. Programs nobody uses are cleared out after 90 days. And the ones that prove themselves can be approved and become certified skills. Four, saved knowledge. The reports, outputs and code snippets that people approve are kept in the knowledge base. So, if someone asked for the same information last week, it's already there. And next to the agent, the SMART KS knowledge system. Here, people review the work and choose what to keep. Yes, this is true. This report is good for everyone. Approved knowledge is saved and indexed right into SMART TS XL as its own data type, so it's found instantly. There are four ways to give the agent a job. From Claude or any LLM connected to the agent. From the prompt box in the SMART KS web app. From a program call, in any language, on any system. A microservice, any app, any process. Or on its own, on a schedule or triggered by an event. Picture this. Someone adds a new field to a database. That triggers SMART KS to check it. Does it follow your naming standards? Is something similar already in use? Or, a JCL job crashes in the middle of the night. The crash triggers SMART KS to investigate. What happened, where it failed and what the error code needs. So by morning, the fix is already on the move. Now, let's watch it work. A developer sits at his desk in Dallas. He opens Claude, connected to the SMART KS agent, and types a huge request. Not a question, a whole project. He clicks send. The request reaches the agent. The agent can work with one LLM or several and pick the best one for each task. One that's great with COBOL, another that's great at documentation or flowcharts. That keeps token costs down and you control which LLMs it uses. First, the agent uses the LLM it chooses to analyze the request. That LLM now knows everything the agent can do. Together, they plan the job. Then they work through it, step by step. Run this certified skill. Use these SMART TS XL tools. Pull this approved knowledge. Reuse the report that was saved last week. Write a quick script in the sandbox and run it, step after step, until the job is done. Here's the key. All of that runs inside the agent, on your code, in SMART TS XL. The certified skills, the tools, the scripts, the AI just wrote, they all run right there. They don't feed your code through the AI. So, they don't cost tokens. When it's all done, the agent wraps it up as one finished answer. The developer in Dallas just gets everything. He never sees how much work or how many programs ran behind it. And his token bill stays small. And if his report is worth keeping, he can publish it to the SMART KS knowledge system for everyone. Next time, anyone who needs the same answer gets it right away. All that work keeps paying off. SMART KS, the knowledge system. One request in, finished work out, mainframe and distributed. Coming soon from IN-COM Data. We started today with one idea. What you say is what you get. You saw what you can do today. What's coming very soon and you were the first to see SMART KS. Together, IN-COM and our clients are ready to use AI in the best possible way. So now it's your turn. Just say it.
