Even zero-second development speeds release only 1.5x: Dingcodingco on why companies want 'product builders'
Nov 7, 2026Original 딩코딩코 · METR · Google DORA ↗
In 3 lines
- Even with zero development time, planning and verification remain, so a three-week release becomes two, only 1.5x faster
- A field account: pull requests up 60% with AI, yet revenue did not follow while incidents and rework rose
- Companies want product builders who clear the choosing and checking stages; for juniors, 'why I chose it' is the edge
"If development took zero seconds starting tomorrow, how much faster would we ship?" That is the opening question of backend developer Dingcodingco's 20-minute version of his LOOP:TALK lecture. His answer: a three-week job becomes two. The bottleneck moved from coding to choosing what to build, and companies now look for product builders who can clear that stretch themselves. We added context from the METR experiment and Google's DORA report.
Zero-second development still only cuts a three-week release to two
Say a feature takes one week to plan, one to build and one to verify and ship. Even if AI cuts development to zero, the total drops from three weeks to two, a 1.5x gain. Dingcodingco said executives spend about $200 per developer per month on AI and expect several times the output, but at his company and others it barely rises. Only the coding step got faster.
It feels much faster: a one-hour ticket estimate now takes five minutes
Comparing his own 2019 and 2026 as a backend developer, he said ticket estimation fell from an hour to five minutes and incident response from two hours to 15. Google's 2025 DORA report found 90% of software professionals use AI and most report higher productivity. In METR's randomized trial, 16 experienced developers expected a 24% speedup but took 19% longer with AI, and still believed they had been about 20% faster afterward. That result reflects early-2025 tools.
METR · METR ↗Widening one stretch of highway to eight lanes does not get you home sooner
Dingcodingco split the path to a product into six stages: finding the problem, choosing a direction, planning and design, building and verifying, release, and checking usage. AI widened the middle ones, planning/design and build/verify. The bottleneck moved to choosing what to build and checking that it matches the direction. DORA calls AI an amplifier of an organization's strengths and weaknesses, and notes individual gains are clearer than organizational ones.
Google DORA · Google ↗Hand the choosing to AI too, and you get a product nobody can explain
In his 10-week backend bootcamp, students with no prior experience built projects such as incident monitoring and detection systems. When asked questions at the demo, they could not answer. Their documents were full of lines like 'under clause X, 32% on 270,000 won'. They had delegated problem discovery and direction to AI. He calls the pile-up of unverified features and decisions 'AI divergence'. A user asking why they got 275 points cannot be answered with a clause number.
Pull requests rose 60%, revenue did not follow
Dingcodingco relayed remarks from Inflearn developer Lee Dong-wook: pull requests and code output are up more than 60% from last year, but revenue did not rise as much, while incidents and rework increased. Developers hand building to AI and QA hands checking to AI, passing the burden along. He called code reviews where each developer's Claude argues for its own work a 'Pokémon battle'.
Two hours of development, six hours of waiting
If an eight-hour task now takes two, developers become part-timers, and the rest of the day goes to waiting for planning decisions. A question like 'approach A takes three days, B takes four but may affect coupons' sends a planner off to renegotiate with another team for days. He says product builders are emerging to cut those round trips, and cited job postings seeking product builders rising from 1.5% to 4.9%.
A product builder is a plunger, and not every stat needs to be a five
A product builder does not replace the PM. He sees companies wanting someone strong in their own role and at levels two to three in neighboring ones, with AI filling the gap. Frontend engineers bring data on where users stop; backend engineers propose bottlenecks and validation metrics from stored data. He described a PM with ten years at Toss and Today's House who gathers code and Jira context, then settles a meeting with 'A is not feasible, so we go with B', halving communication costs.
Speed of judgment over years of experience, and a chance for juniors
He said companies want judgment speed layered on top of implementation skill. Seniors who only deliver what they are told on schedule can become bottlenecks, which looks like an opening for juniors. But narrowing options requires someone who has built something at least once. He urged portfolios to show why you chose something, not just what you built. In his bootcamp, questions shifted from 'how do I build this' in weeks one and two to 'which of A and B' by weeks five and six.
Three questions to ask your current project
The talk ended with three questions. Whose situation does this improve? Is there a smaller way to check first? What will tell you after launch? He said asking them reveals the blocked stretch, and the person who clears it is the product builder. His sample line packed problem, choice and metric into one sentence: noticed teammates missing announcements, fixed notifications, tracked read rates in logs for a week.
