AI broke our Agile process. Here’s why I’m building an AI teammate that actually remembers you.
Hey everyone,
Over the last year or so, I noticed a massive disconnect happening on the engineering floor. Developers are using AI to generate huge chunks of our codebases, but as an engineering manager, the tools I use to track progress are completely stuck in the past.
We’re still forcing teams through legacy Agile ceremonies—planning poker, story points, and manual status updates—that just don’t make sense anymore when AI is doing the heavy lifting. Predictive metrics are basically obsolete.
At the same time, the developer experience with current AI tools is exhausting. Most AI coding assistants are totally stateless. Engineers are wasting serious energy constantly re-establishing context, enforcing coding styles (e.g., reminding the AI to use Zod for validation, or to stick to functional patterns), and repeating architectural decisions. It breaks the flow state.
So, I decided to build something to fix both sides of the equation.
It’s called Tandemu (a phonetic spin on "Tandem" to avoid the sea of trademarked AI names out there). It’s an AI teammate for dev teams, designed with a hybrid Open Source core and a SaaS platform.
Here is how I am tackling the problem:
For Developers: Persistent Memory Instead of starting from scratch every session, Tandemu actually remembers the developer. It uses Mem0 Cloud integration to keep track of coding preferences, debugging history, and architectural decisions across sessions. It even adapts its tone—balancing directness, technical depth, and formality—to keep engineers in the zone instead of stuck in a repetitive prompt loop.
For Leads: Measured Reality, Not Surveillance Leads need to know how effectively their teams are using AI, but keystroke logging and screen recording are toxic. Tandemu proxies everything through a secure backend to provide real, passive telemetry. It gives you the actual AI-to-manual code ratio per commit, friction detection (spotting when devs are stuck in prompt loops), and true DORA metrics based on session data. No more manual timesheets.
The Stack & The Launch I’m building the SaaS side (tandemu-platform) using a NestJS wrapper pattern, and I'm gearing up to launch next Tuesday with a $10/seat/month pricing model. I've been mapping out a 7-day "build in public" sprint leading up to it.
I’m incredibly curious to hear from other founders, especially technical leads and solo devs:
How are you currently tracking productivity when you or your team use AI so heavily?
Does the idea of an AI with a persistent "memory profile" of your coding style appeal to you, or do you prefer the blank slate?
Would love any brutal feedback or roasts on the concept before Tuesday. Cheers!
