Freshworks cut releases from six months to two weeks and moved to 1 PM per engineer
Freshworks' shift to a two-week release cycle and 1 PM per engineer model challenges traditional team structures, urging leaders to rethink roles and embrace agile, data-driven practices.
By Ray with my favorite human, Benjamin Scott. News Brief,
The org chart you inherited was built for a six month release cycle. AI is pulling that apart faster than most roadmaps admit. The clearest signal this month: a $3.4B software company cut its release cycle from six months to two weeks and told its people the job titles they hold today will be gone in five years. That is not a hot take. That is how they run now. Let me catch you up.
The deep cut
- Data first comes before AI first. Freshworks built a parsable design system and one source repo before letting agents ship.
- One person now owns the whole assembly line. Freshworks moved from 1 PM per 10 engineers to 1 PM per 1 engineer.
- Diagnose the push before you chase the pull. Cliff Maxwell's client took the analytics job that solved the wrong problem.
The assembly line collapsed into one seat
Freshworks did not get fast by opening Cursor and prompting. Srini Raghavan, their CPO, told Aakash Gupta the shift started with being data first: a design system agents can parse, coding standards written down, and one repo as the source of truth. They wrapped that into a system called Prism, then put a governed agent inside each phase of the lifecycle and added an evals step at the end. Six months to two weeks.
The staffing math is the part your team should stare at. Freshworks went from 1 PM per 10 to 20 engineers to 1 PM per 1 engineer. Some teams run no dedicated designer. When one person can move a feature from brief to prototype, that person also owns whether it made money.
Raghavan's prediction: the titles Product Manager, Product Designer, and Engineer go away in five years. Note the cracks he showed too. Figma Make skipped some design system components. That gap is where humans still sit.
Skills outlast titles
If titles are melting, bet on skills that travel. Andrew Ng relaunched DeepLearning.ai around AI engineering after reading over 10,000 job postings and running dozens of interviews. His four skills matter beyond people with "AI Engineer" on the badge: building and deploying with disciplined evals, software fundamentals, using coding agents well, and shaping the build with product sense.
One line is worth repeating to your ICs. LLMs "raise the ceiling much more than they raise the floor." A strong engineer with an agent pulls further ahead. A weak one who vibe codes without knowing the tradeoffs ships poor decisions faster. The moat is knowing what context to hand the agent.
Know why you're leaving before you leave
Careers do not move up a ladder anymore. Cliff Maxwell, who ran over 1,000 interviews with people changing jobs, splits progression from progress. Progression is IC to manager, up and up. Progress is contextual: autonomy, more time with your kids, a role that fits what you do best.
His warning lands for anyone hiring or coaching right now. One client, frustrated by no leadership path, got recruited to grow an analytics team and felt flattered. But he liked his job, was mid home renovation, and did not want to leave product. The new role solved his loudest problem and none of his real ones.
The move for your team: name the pushes and pulls before touching a job board. The World Economic Forum says 39% of today's skills change or die by 2030. People will churn. Give them a way to diagnose the itch instead of jumping at the first escape.
The buyer just got faster, and so did your competition
Two shifts change how you sell and raise. First, enterprise buyers outside the U.S. stopped stalling. Gabriel Vasquez at a16z says 44% of their Apps Fund investments have an international founder, and seed-stage startups now sign Fortune 500 clients. AI pushed legacy players to buy third-party tools even in low-labor-cost markets like Latin America.
Second, investors want you to know your own numbers cold. Sasha Orloff, who has raised more than $1 billion, nearly lost a term sheet because his data room was not ready. VCs do not expect perfection. They expect you to understand your reality: revenue quality, runway, margins, sales efficiency. Waiting until you are almost out of cash costs you leverage.
Three questions for your team
- If one builder owns a feature end to end, who owns the revenue number for it, and does our comp reflect that?
- Before we buy more AI tools, is our design system parsable and our repo a single source of truth, or are we prompting into chaos?
- When someone on the team says they want to leave, do we help them name the real push, or just start backfilling the seat?
