Season's Savings | ₹5/mo for 3 months

₹199/month after first 3 months. Renews automatically. Terms apply.
Leapfrog cover art

Leapfrog

Leapfrog

Written by: Hugo Lerias
Listen for free

About this listen
Most enterprise AI doesn't fail on the model. It fails on the engineering around it: integration, data, deployment, evaluation, cost, security. That's a cloud-engineering problem, which means the engineers you already have are the ones most ready to deliver it. Each episode takes one chapter of the free field guide Leapfrog and talks it through in under ten minutes. Every one ends with three things you can go and run. Book and code free at leapfrog.lerias.org. My other show, on good work that never surfaces in big organisations: Below the Approval Line. Personal views, not my employer's.Hugo Lerias
Episodes
  • The Four Reasonable Objections
    Nov 1 2026

    Somebody proposes building something small with AI. Four sentences come back: yes, but the migration will take years; yes, but we have contracts until 2029; yes, but security hasn't signed off; yes, but let's wait for the platform team. None of them is wrong — and that's exactly what makes them dangerous, because a room full of reasonable objections produces the same result as a room full of bad ones.


    This episode is about why that sentence used to protect you and now doesn't. Cloud was a should; AI is a must, because the job itself is being redefined around it. So deferring in step with your competitors was safe, and deferring while some of them go straight past you is not.


    Then the numbers. Inference costs fell more than 280-fold in two years. Open-weight models closed the gap with the best closed ones from eight percent to under two. And yet roughly 95% of organisations investing in generative AI see nothing on the income statement, and over 80% of projects never reach production.


    Read the post-mortems and the cause is almost never the model. It's integration, data, deployment, cost control, security — engineering problems, which is the discipline you've been practising your whole career. The failure rate isn't a warning to stay away. It's a map of where the value is stuck.


    Run it yourself — the labs for this episode:

    • Lab 01 — make one real API call and read the meter, token by token

    • Lab 10 — give your model six tasks that look equally easy, run each three times, and watch it be inconsistent

    • And one with no code at

    Show More Show Less
    12 mins
adbl_web_anon_alc_button_suppression_t1
No reviews yet