Skip to content
FindGlocal
Code Scientists

Code Scientists

3619 South Avenue, Springfield Software Company

We are product development experts! At Code Scientists, we are specialists in the area of commercial software development.

Working with a variety of technology stacks, we take your idea from concept to MVP to help you support it after release, building AI, mobile, web, and desktop applications with the cloud infrastructures to support them. Having spent many years in the commercial software industry, working on both business and consumer applications, we are an excellent resource for startups, businesses wanting to automate "heavy thinking", and software companies that need solid solutions fast.

11/02/2026

We have discovered that our ai-governance model is a little bit too strict. They didn't stay blocked long; however, last night they "fault stopped" because the engineer agent performed an unlogged web fetch of Firebase documentation after we gave Claude instructions to review ChatGPT Sol's comments on their first webapi pull request.

1️⃣ The engineer took the "easier to ask forgiveness than to get permission" approach.
2️⃣ Afterward, the engineer reported the infraction to the Claude lead.
3️⃣ Though we don't want to get blocked on this, the bounded-autonomy worked.

We are going to loosen the reins a little bit for sprint 5. But, it's nice to see that the agents are working within the authorization framework that we have created (truly a collaboration between human and artificial intelligence). Even though the engineer executed the command, it saw its mistake, knew not to use the results, and the system entered a fault stop state until we could intervene.

🧑‍🔬Experiment Overview 🤖
This experiment asks whether a pool of AI agents, orchestration rules, automated tests, and human review can operate as a trustworthy software sprint team. The immediate product context is Say it with Simon, an AAC application from Software for Simon. The broader objective is a repeatable operating model for AI-assisted software delivery.

10/27/2026

For the past 6 1/2 weeks, we have been experimenting with ChatGPT Sol Medium, Claude in experimental team mode, and bounded-autonomy. The purpose is to see if we can interact with the agents just like we would any other sprint team. We work together almost every day with the AI agents, observing the typical SCRUM rituals (though our modified version of the process).

Garry Tan (in April) bragged that he was shipping 37,000 lines of AI-generated code per day. A senior software developer reviewed his code and found many severe shortcomings. Naturally, Garry was subsequently roasted on social media. (See https://lnkd.in/gNhsKud2.)

The takeaway? Slow the heck down!

With our automated sprint team, the guidance and authorization protocol limits the number of active agents on the Claude-side to five. Every pull request is reviewed by a human, ChatGPT Sol Medium, and at least one Claude reviewer acting in the agent role. We're not generating 37,000 lines of code per day, and don't want to. That's too much to keep up with. Instead, we're executing planned sprints that we could execute in a day or so versus two weeks for a typical five-person sprint team.

We feel like this important. Instead of running "balls out", which is a real engineering term by the way😃, and producing code that's unmanageable and full of holes, we're operating at a sustainable rate and 10x the velocity of a human team.

And, so far, the "team" is producing code just as good as or better than many of the human teams we've led or participated in.

10/22/2026

Generative AI tools experience pattern drift over time, requiring periodic realignment with the established patterns within a project. The phenomenon doesn't appear to be related to the type of content you're generating, whether you're generating text, images, or code. (Images are the most difficult to repair.)

We use generative AI every day to generate documentation and write code. It saves us a ton of typing. There is still a lot of development thinking that goes on around this process. We've got to make sure that what's generated adequately solves or documents the problems I'm working on.

We also have to make sure that the models use patterns consistently. Using consistent patterns improves maintainability and quality.

Over the past year and a half using these tools nearly every day, we have noticed that they drift, especially on larger projects. The models may start out writing code using one pattern, such as a state management pattern in a mobile app, then suddenly switch to some other pattern, going from Riverpod to Provider in Flutter. Another example is going from one class definition per file to multiple class definitions per file while working on the same project over time.

Today, we noticed this same drift in documentation. Our ChatGPT "scrum master" generated experiment documentation for sprint 3 of the automated sprint team experiment using a totally new format, even though it had helped create the documentation templates. Same chat. Same context. It has access to the templates.

In tools like Cursor, the pattern drift may be due to model selection, since we use model and model strength set to automatic unless we need a "special set of skills". In ChatGPT's case, maybe it's caused by context abbreviation or compaction.

Regardless, it's a huge pain in the ass. 🙂

⬇️ Experimental Overview ⬇️
This experiment asks whether a pool of AI agents, orchestration rules, automated tests, and human review can operate as a trustworthy software sprint team. The immediate product context is Say it with Simon, an AAC application from Software for Simon. The broader objective is a repeatable operating model for AI-assisted software delivery.

The experiment is not limited to code-generation speed. It evaluates whether agents can work within authority, coordinate safely, identify genuine product decisions, review one another critically, and produce maintainable software while leaving final control with a human product owner.

10/20/2026

"Blanks" aren't necessarily blank. 🙂

Any of you who have internationalized a product know this already. In fact, there are about 4,258 ways that a "blank" line may not actually be blank. For the users of an AAC (Augmentative and Alternative Communication) device, this could mean creating a tile for a word that appears broken or possibly unfixable.

This is what the automated sprint team decided to address for sprint 3 of the automated product development experiment. So, they proposed and executed the following:

☑️ Defined "visibly blank" using Unicode 18.
☑️ Added code to prevent saving non-visible words.
☑️ Added edit-mode indicators (their decision) for broken words.
☑️ Added extensive testing to preserve the rules against future refactors.
☑️ Found one additional gap during PR review not covered by the tests.

The final bug was caught by ChatGPT Sol Medium, the "scrum master" of the experiment. Having multiple models from different platforms work together is proving to be very beneficial, in my opinion.

🔬Experiment Overview 🔬
This experiment asks whether a pool of AI agents, orchestration rules, automated tests, and human review can operate as a trustworthy software sprint team. The immediate product context is Say it with Simon, an AAC application from Software for Simon. The broader objective is a repeatable operating model for AI-assisted software delivery.

The experiment is not limited to code-generation speed. It evaluates whether agents can work within authority, coordinate safely, identify genuine product decisions, review one another critically, and produce maintainable software while leaving final control with a human product owner.

10/16/2026

The most interesting bug our automated sprint team found this week wasn’t in the code. It was in the authority we gave the team.

We are experimenting with AI agents in familiar roles: scrum master, engineer, auditor, and independent reviewer. During one review, an agent caught two subtle problems:

🪲 A rule intended for two small tasks accidentally constrained the rest of the sprint.
🪳 An estimate was written as though it were a measurement.

Neither problem would fail a unit test, but both could affect how the team operated and how we interpreted the evidence. The reviewer refused to approve the work. The language was corrected and independently checked again before anything was committed.

It is an early observation, not a broad conclusion, but the lesson is compelling: AI review diversity may matter because software projects contain more than code. Requirements, permissions, evidence claims, and governance language can also contain defects.

10/14/2026

Our Chief Scientist, Markus Pope, was somewhat scolded this week by the Claude part of the experimental sprint team, as well as Chad (ChatGPT). "They" discovered that someone had run a 'git fetch' in the repo they're working in, and they wanted to know who it was, why it was done, and document the provenance of the change. 🙂

We're considering that a successful result in the bounded-autonomy experiment, though it may be a bit pedantic. We've all worked with that team member, haven't we? 😀 (You know who you are.)

Automated Sprint Team Background →
This experiment asks whether a pool of AI agents, orchestration rules, automated tests, and human review can operate as a trustworthy software sprint team. The immediate product context is Say it with Simon, an AAC application from Software for Simon. The broader objective is a repeatable operating model for AI-assisted software delivery.

The experiment is not limited to code-generation speed. It evaluates whether agents can work within authority, coordinate safely, identify genuine product decisions, review one another critically, and produce maintainable software while leaving final control with a human product owner.

10/06/2026

Last week, we deployed changes to production for a customer, and it was completely uneventful. That's what you should strive for. What tools you use to get your work done matters significantly less than the *diligence* with which you do your work.

Whether you use AI or not:

📒 Plan your changes.
🔍 Scrutinize your plans.
🛠️ Implement the plans.
📲 Test and review your work.
🧐 Welcome peer review and criticism.
📝 Create and test release playbooks for every release.

Your releases won't always be uneventful; however, you will experience less stress and lower your risk by following this simple process.

10/04/2026

Our experiment in bounded-autonomy progresses. Last night, ChatGPT and the Claude team worked through some of the differences between user and talker authorization for the AAC we're building. This is a design only sprint.

We had to clarify some behavior, which resulted in new work. Claude wanted to add it to the current sprint, while ChatGPT (Chad) correctly pushed it to Sprint 3. During authorization and closeout, Claude indicated that the content would push us over the token budget.

We delegated the remediation to Chad, and he chose to reformat the context to pull it back under our 15,000 token limit. We were curious to see what direction it would go...Chad chose what we had already determined to be the proper course of action, reaffirming that we are in alignment with regard to sprint team and project management.

10/02/2026

We've created an automated sprint team using ChatGPT and Claude. ChatGPT (Chad) acts as the scrum master, while Claude (in team mode) acts as the development team.

Most significant finding in the Automated Sprint Team experiment from Sprint 2 ex*****on last night:

ChatGPT proactively reviewed a pull request through previously granted access that I had forgotten I had given it.

Claude reported that PR #10 was ready for human review and merge. Acting as scrum master, Codex independently opened the live PR, inspected its metadata, commit, complete diff, attribution language, and CI result without being explicitly asked to perform the review.

🔒 Bounded-autonomy) 🔒
The action was read-only. Codex did not approve, modify, or merge the PR. I approved the initiative after the fact and reaffirmed Codex’s scrum master role.

This demonstrates useful delegated initiative emerging from previously granted tool access. The agent recognized a review opportunity, acted within a safe read-only boundary, and reduced the human’s verification burden.

Wow.

02/09/2026

The secret to actually moving the needle isn't found in a giant leap. It's found in small, manageable steps that make what feels impossible, begin to feel possible. Today, we’re just focusing on taking the first step. Whatever big project you've been putting off, or have felt apprehensive to start, break it down today. What is the one small thing you can do right now to get started?

Other Software Companies in Springfield

Show All
  • Purpose Connect

    Purpose Connect

    405 N Jefferson Avenue Suite 1005, Springfield 65806

    Hire international talent with confidence—A better workflow for every stage of the journey | Backed by Pharrell Williams’ Black Ambition 🚀

  • HouseKeeper

    HouseKeeper

    2343 N Delaware Avenue, Springfield 65803

    The world's most efficient short-term rental turnover software

  • Fleet Technology Services

    Fleet Technology Services

    4620 W. Chestnut Expressway, Springfield 65802

    Fleet Technology, Inc. provides technology software and service solutions for the transportation industry including advanced shop maintenance software, safety and monitoring for in...

  • ConGenius

    ConGenius

    3107 E Chestnut Expressway, Suite G, Springfield 65802

    If you're a contractor, you know how hard it is to run a business. Our goal is to help you maximize your time and impact. We're a construction software company built by contractors...

  • App Developer Magazine

    App Developer Magazine

    Springfield 65802

    App Developer Magazine is a monthly publication dedicated to the app developer and app entrepreneur.

  • Gabbl AI

    Gabbl AI

    2131 W Republic Road, Springfield 65807

    Your Intelligent Chatbot Platform

  • Transform customer experience with virtual queue management. Help Springfield businesses serve 30% more customers daily with zero additional staff. Join the waitlist revolution! 🚀

  • RuntWerkx Systems

    RuntWerkx Systems

    1126 S Colgate Avenue, Springfield 65802

    RuntWerkx Systems builds modern industrial software designed to streamline operations.

  • ZAG Web Development

    ZAG Web Development

    Springfield

    ZAG Development is a locally-based web development startup founded by three students.

Alerts

Be the first to know and let us send you an email when Code Scientists posts news and promotions. Your email address will not be used for any other purpose, and you can unsubscribe at any time.

Subscribe

We will notify you when anything happens in Springfield.