Faster Code Isn't the Point

The GetDX Engineering Leadership Summit brought together engineering leaders from Airbnb, Vanguard, Intercom, Twilio, 1Password, and others. Much of what came out of it cuts against how most companies are currently thinking about AI.

The bottleneck was never writing code

Microsoft's research is straightforward: engineers spend about 14% of their time writing code. The other 86% goes to code review, testing, documentation, story refinement, and deployment. AI tools handle the 14% well. The 86% is largely untouched.

For Libyan companies considering AI investment, this matters directly. Buying an AI coding tool is the easy part. Whether your delivery process can absorb faster code is the real question. In most cases it can't, and that's true globally.

Vanguard found that engineers writing code 30% faster had zero effect on end-to-end cycle time. PMs were still taking 3-5 days to write stories. QA was still running manual tests at sprint end. Faster code just meant developers waited longer at the next queue.

One Vanguard project made this concrete: an agent finished implementation in 2 days, then waited through a 4-day design review, a 3-day API onboarding process, a 5-day security review, and a 2-day deployment approval. The fastest path to ROI was fixing the security review queue, not improving the model.

AI made the existing bottlenecks visible. Code review was always the constraint. Meeting overhead was always the constraint. AI removed the buffer that was hiding them.

Training matters more than which tool you pick

Indeed put around 2,000 engineers through a structured, tool-agnostic AI course. Those who completed it cut coding time by 36%. The differentiator was not the tool. It was whether people knew how to use it well.

LeptisCode works with Libyan businesses three ways: training engineering teams to use AI tools at a professional level, building applications and AI-powered products end to end, and augmenting existing teams with senior Libyan engineers held to international standards. The research above applies to all three. Whether you need your people upskilled, a product built, or extra engineering capacity, the underlying issue is the same. AI only delivers if the people and processes around it are ready.

Microsoft observed that under time pressure, most engineers revert to single-threaded pairing-style AI use rather than running multiple agents in parallel. The tooling moved faster than the behaviour. Give someone a power tool and they use it like a hand saw.

The engineers are capable. What Libyan teams need is structured practice, applied to the workflows they actually operate in.

Daily usage is what moves the numbers

Airbnb segmented engineers by hours of daily AI use. Engineers using it four or more hours a day more than doubled their output versus pre-AI baselines. Occasional users saw modest gains.

Airbnb runs at 65% higher throughput and 59% AI-authored code with no mandate in place. Intercom has 95.9% of pull requests authored by Claude, throughput doubled in 9 months, and defect backlog down over 50%. These results come from teams using AI seriously, every day, on real work.

Token spend is the new cloud bill

One company of 300-400 engineers hit $128,000 a week in token spend. For Libyan businesses with tighter budgets, the lesson from 1Password is worth taking: the AI token bill needs the same rigour as any infrastructure cost. Most organisations default to the most powerful models for tasks a cheaper model handles fine. Pick the right model for the right task and build cost discipline from the start.

The unit of productivity is the team, not the engineer

Airbnb has twice as many AI users as engineers. PMs, designers, and data scientists use AI coding tools daily. Mercari saw a 45.7% drop in accounting tasks and a 60% reduction in help desk workload from AI tools built by non-engineering teams.

AI adoption is not an IT project. Companies that pull ahead treat every function, operations, finance, HR, customer service, as part of the change. The ones that treat it as an engineering initiative will get engineering-sized results.

---

AI amplifies whatever you already have. Slow process, bad structure: AI makes it worse. Daily usage, proper training, clear standards: the gains compound. The Libyan businesses that move now will be ahead of those that wait. The tools are available. The question is whether the people using them know how.

To discuss what this looks like for your team, reach out at hello@leptiscode.com or visit leptiscode.com.