Tekin Method Take the test

Decision Architecture for Salesforce and Agentforce

Your system is making decisions. Can you see them?

Before an agent goes live, someone has to decide what it may decide. In most companies nobody has, so the builder decides. I find those decisions, give each one an owner, and hold them in Salesforce where the agent cannot talk its way around them.

Six languages. Nothing is saved or sent.
Tested in my own org: 7 of 7 acceptance tests passed. 10 of 12 ways around the agent still worked.

Where this stands

It isn't proven yet. Nobody has proven the alternative either: most agent projects run on rules nobody decided.

Decision Architecture is a working theory and Tekin Method is a young practice. I am looking for the first five companies to test it with, in the open: free for them, done by hand, with a written result they keep. If it doesn't help, you'll know quickly and so will I.

How I work

Six steps, the same on every engagement.

The work is semantic before it is technical: what the words mean is decided before anything is built.

  1. Question the storyWho, what, when, how and why, until the answers hold. Every gap becomes a named decision with an owner.
  2. Separate authority from capabilityWho the actor is, what they are allowed to do, and what the system lets them do.
  3. Design the work before the toolEntry criteria, conditions, exceptions, calculations, impact, order of execution.
  4. Choose the tool on evidenceFlow or Apex, compared openly against the team's capacity.
  5. Make it durableCheckpoints, edge cases, a trace the organization can learn from, intervention points.
  6. Ship in atomic unitsMetadata packages in a dependency-aware sequence.

My story

I kept seeing the same failure.

I have spent more than eight years delivering Salesforce, mostly in regulated industries: four and a half years as an architect at EPAM Systems (PolSource), four years at Intech Consulting, and four years in SAP CRM consulting before that. I hold 12 Salesforce certifications.

Projects rarely failed in the build. They failed on rules nobody had decided, which ended up decided by whoever held the keyboard. With agents that gap got faster and quieter. In 2026 I went independent to work on that layer. I live in Istanbul and work remotely.

Contact

If one of your processes came to mind, write to me.

Send me the sentence the test gave you, or tell me where it got your process wrong. I answer every message myself.