Pulse Framework

How to ship faster and automate your dev process with AI

✨

This framework lists actionable steps on how to integrate AI agents into your project development process and measure success of your efforts. It’ll be helpful for any founder and project leader who’s running the existing dev team from 1 to 20 people.

Today it’s not enough to apply powerful AI agents such as Claude Code or Codex just to write code. When building a real project with real customers and risk, there is much more to “building" than it is for MVPs.

You need to apply AI agents on all phases of your SDLC (software development lifecycle).

SDLC (software development lifecycle) phases and metrics

Phase	Metrics and data	Assets to create	Tools
Plan	• Tasks ready to be built	• Backlog	• Linear (Jira, Clickup)
Design	• Prototypes created	• Clickable web prototypes	• Claude Code • v0, Lovable
Build	• Commits • Pull requests	• AGENTS.md • AI Skills files • AI project docs	• Claude Code / Codex / Cursor • AI skills and plugins
Test	• Pull request review time • Test coverage %	• Automated tests • Automated code reviews	• Greptile • CodeRabbit
Release	• Release/deploys frequency • Tasks released	• Changelogs • Help Center	• Fin • Mintlify • Featurebase
Operate	• Support chat messages • Production bugs, errors, logs • Customer satisfaction rate • Meetings follow-ups and transcripts	• App Analytics • Automated error handling • Customer chats • SOPs • Daily updates with images/videos	• Sentry • Axiom • BetterStack

Tasks

Tasks and reports are often underlooked, but the way how you define tasks and their attributes greatly changes the outcomes of your work. Spend time to organise dev process.

Task Status

Backlog

Sprint

In progress

Under review

Ready to test

Testing

Ready to release

Released

Tasks are 3 categories: Features | Improvements | Bugs.

Tasks should be broke down and estimated in hours, not days. No tasks should take longer than 40 hours (ideally up to 16-24 hours).

Tasks should be useful for the end customer and should not turn into development-specific challenges.

Make sure backlog is always filled with tasks. Automate task sourcing and creation (through customer support and automated monitoring).

Make roadmap easy to digest and update (it’s always buried). Roadmaps are often more convenient in Google Sheets, don’t rush to automate.

Async work

Async work means work through proactive written communication rather than constant meetings. Async work enables focused work with the global team in different timezones. Aim to automate repeated work through published assets, processes or tools.

Stimulate rituals:

Rituals make it possible to measure metrics and align the team around releases:

Daily updates in Slack

Daily commits and published assets

Weekly releases to prod

Replace meetings with published assets:

Daily team updates [instead of daily meetings]

Assets: screenshots, video/loom demos, docs [instead of random calls]

Changelogs [instead of weekly status]

Help center docs [instead of customer calls]

Make work visible through transparency:

Brainstorms in shared Slack channels

Reviews on GitHub

Discussing tasks inside task board (Linear/Jira)

Document all the decisions (follow-ups after the calls and assets instead of calls)

Team

We see typical production project teams still inherit traditional team composition, e.g.: Engineer, Designer, QA, Product Manager, Product Marketer.

With AI you have more room to combine roles, which will also speed up handoff between people. In early-stage startups, all roles are handled by 1 or 2 people. Aim to keep teams small, up to 3 people and break down larger teams to smaller squads.

Train team to work with AI:

Prefer published assets and proactive updates rather than calls

Every person should present their work themselves, learn how to do it well, dedicate time for presentation

Assign owners on tasks and assets so it’s clear who’s responsible for the end result

Be empathetic to your team. Numbers and data is great, but keep in mind that people take time to adopt to new rules, and charts don't always show a full picture. Focus on promoting best performers.