FileMaker Agentic Development

ai generated image with agent at your laptop, and ai generated error . . .

The role of the 'problem solver' is to provide people what they need, not what they want . . .

Take a Poll at a Town Hall Meeting - filled with Claris FileMaker Developers - what they want to see next for the Claris Platform, and they'd vote for an 'AI Chat' window 'within' the FileMaker App, that interacts directly to their FileMaker database - with the ability to update elements, tables, table occurrences, fields, calculations, scripts, even call plug-in functions.

Sure, the result would allow for a 'natural language' approach to code creation, as per the functions we have now for end-users creating ad-hoc reports, and querying data based on a more agentic interface, rather than deterministic coding.

This is only one small part of the Development Process, and the Developer Experience.

I'm suggesting Claris FileMaker Developers need to go beyond our platform - to acquire knowledge and skills in this new exciting coding ecosystem - to stay valid.

Claris FileMaker Developer Preview

Claris recently presented their vision of the future for Agentic Coding in FileMaker that will arrive in the Summer this year - of course, Winter - for those of us here Down Under.

Richard Carlton followed up with a FileMaker Training Video with Rick Kalman, along with Ernest and Todd from Proof regarding ProofKit.

Claris' vision keeps the Agent 'external' to the platform - so not within the FileMaker App, and more to the frameworks currently provided by a number of Claris Partners.

Claris FileMaker 26 delivers the 'under the hood' components for this shift towards instructing the Agent what you require, as the Orchestrator of the Agent(s).

Agentic-FM

One of the latest toolsets for the Claris FileMaker Developer is Agentic-FM.

Matt has been crafting this framework to provide a number of moving parts for Agentic Coding in FileMaker, along with an Open Source solution and extensive documentation.

I've spent many hours within this framework with real world projects covering a range of small changes to an existing system. This included functions re-write of a key section of a field, moving a function to a different part of the workflow, and also a tool to move PDFs between Server Folders based on their reference within a FileMaker record.

AG-FM Plugin.

We've all installed and configured FileMaker Plugins to extend the features and functions available to the Developer - the centre of this is the AI Chat window.

Highly configurable, the plug-in has access to read the schema and create new element directly in the open FileMaker Database.

This AG-FM Plugin is the perfect 'starting' place for Developers on their Agentic Journey.

AG-FM Unplugged.

For those who have answered the agency to Code with Agents - this set of tools allows the Developer to work within their preferred Harness.

In my case, Claude Code - with the ability to read the schema, review and update code within the workflow of the FileMaker Solution, and to either create instruction sets to update, or to reach out and create or update scripts.

This is for those of us who want to look under the hood, and move beyond the basics.

Agentic Developer Experience (AG-DX)

Think of the Agentic Developer Experience as working with a newly appointed intern that needs your experience to ensure the end result is what you expected.

Share the Requirement with the Agent on a new project or task - this is currently no different to an Intern or Junior Developer.

And just like an Intern - the Solution, Code and anything else - needs to be double-checked, as this is not a one shot solution - this is no different to existing Development Methodologies and Processes.

Consultant vs Development Workflow

As Consultants - and yes, I'm making that switch from 'Developer' deliberately - cutting the code is only one part of what we do. The real work sits upstream.

Let's be clear. No amount of Agentic Coding skips the work that happens before and after the code.

Consider this quick list of steps required for any software project

• Identify the Stakeholders

• Define the Need

• Agree the Success Criteria

• Design the Solution

• Build and Test

• Deploy and Train

• Review and Iterate

The Agentic Workflow doesn't replace this. It accelerates it.

Don't believe the hype on the internet that this Agentic Coding is all about setting up a Mac mini with OpenClaw and you explain your project in one prompt, and that's all done.

In Agentic terms, this is why we now have the Title of 'Orchestrator'.

Agentic Coding Method

Even after a number of successful projects - with your preferred harness configured with skills crafted to your coding style - it's tempting to simply give the Agent licence to update the code directly, as the outcome of the initial prompt.

There's a method of reading the outcome of the prompt, to make the decision to paste the code into the system - and then re-read that the code pasted correctly.

As the Orchestrator, a key process to prompt once, read twice . . .

Developer Cognitive Load

And yes, Developer Cognitive Load is real, and Agentic Coding can lift the load, and move the focus to the design of the code, rather than the code itself.

It's worth considering this advantage and trialing the change. This will feel familiar to those who have mentored or worked with an intern, giving the Agent just enough brief and expectation of output, as we do within any Agile Workflow.

AI Vocabulary

To finish off on a light hearted but related subject - with any new technology - there's a new set of buzz-words and acronyms to learn and impress your non-technical friends and foes.

Agentic.

Artificial Intelligence systems that can autonomously set goals, make plans, and execute multi-step tasks, write code in dynamic environments with minimal human oversight.

Deterministic.

Feed the system the exact same input, it will always produce the exact same output, running through the same sequence of states. There is no randomness, unpredictability, or variance.

Natural Language.

The ability of computer systems to understand, interpret, and generate human language—such as English, Spanish, or Mandarin—to perform tasks

Orchestration.

Directs multiple automated tasks across separate environments, accounting for logic, dependencies, scheduling, and errors, etc.

Harness.

The Wrapper around an AI's LLM, applying the necessary tools, memory, and guardrails to safely control the Agent and complete complex tasks successfully.

Skill.

A structured set of reference files and instructions that tells the Agent how and what to do to ensure the 'context' is understood and the outcome does not hallucinate.

Vibe Coding.

This is not that different to downloading the trial of FileMaker Pro and building Solutions without an coding experience - where most of us got our start.

AI Slop.

Output from a generative AI system that favours quantity over quality.

AI Hallucination.

Where the Agent simply lies, rather than saying it does not know.

Temperature.

AI parameters controlling generation randomness, hardware conditions of dev machines or servers, or a metaphor for team morale.

Linter.

A linter is an automated tool that performs static code analysis to identify programming errors, stylistic bugs, and suspicious constructs without actually executing the code.

AI Psychosis.

Vulnerable individuals develop or amplify severe delusions, paranoia, or altered realities through prolonged, immersive interactions with AI chatbots.

Bias.

The simplifying assumptions made by the model to make the target function easier to approximate.

Domain Knowledge.

The deep and real world understanding of the specific industry or business area for which you are building software.

Taste.

The disciplined judgment to know what the right solution looks like before writing code, and the ability to choose an approach that actually 'fits' a problem rather than one that merely compiles.

Takeaway

Give them what they need . . .

Continue this subject with my next blog post - Filemaker Developer AI- Coding Skill