AI/ES/25/11 Code/LEAD/20 Thu/1005 2026 year ide died steve yegge gene kim amp sourcegraph

• 10:05am - 10:24am Talk: 2026: The Year the IDE Died

  • 
  • Person/Steve Yegge Twitter LinkedIn - Engineering Leader, Sourcegraph/Amp
  • Person/Gene Kim Twitter LinkedIn Website - Author & Researcher, IT Revolution

  • Talk: 2026: The Year the IDE Died 01:10:31 - Steve Yegge & Gene Kim Begin Their High-Energy "Vibe Coding" Talk

    • As AI has grown more capable, software developers around the world have lagged behind the technology advances, and have consistently eschewed the most powerful tools. In this talk I explore why devs are staying 9-12 months behind the AI curve. I'll share a preview of what 2026's AI coding tools will be like, and paint a vision of where we go from here.

  • Person/Steve Yegge

    • AI Developer Trajectory — timeline layout

    • A horizontal timeline with four numbered stations (1–4) spaced evenly across.
    • 2023 (Step 1) sits above the first marker: "code completions (GitHub/CoPilot)." Visual cue: early, narrow capability.
    • 2024 (Step 2) under the second marker: "chat-based programming (CursorAI, Windsurf, Cody)." Visual cue: dialog-driven coding replacing autocomplete.
    • 2025 (Step 3) above the third marker: "agentic coding assistants (Claude Code, Codex, Cline, Amp, Gemini CLI, …)." Visual cue: richer toolchains with multi-step autonomy.
    • 2026 (Step 4) by the final marker: “???” signalling an open frontier.
    • Caption beneath: "Agentic coding is not the Final Form." Emphasis: next phase = orchestrators, implying multi-agent coordination rather than single-agent assistance.

    • Session context

    • Title: 2026: The Year the IDE Died.
    • Framing: devs lag 9–12 months behind AI progress; 2026 tools shift from assistants to fully orchestrated coding systems.

    • Coding Agents as Power Tools — conceptual layout

    • Slide split left/right. Left block: analogy. Right block: limitation + need for evolution.
    • Left side: Coding agents framed as manual power tools (drills, forklifts). Visual cue: interchangeable attachments → “an agent can be many tools.” Emphasis on skill: anyone can dabble, few master.
    • Right side: Manual tools can build “anything” only until the project becomes too large/complex. Modern software exceeds individual capacity, even with AI. Closing line gestures toward industrial automation: move from power tools → CNC machines.
    • Your gloss fits: Claude Code ≈ electric drill/saw — powerful but still fundamentally manual.

    • The Problem: Adoption — two-tile comparison

    • Slide foreground shows two large rounded rectangles side-by-side.
      • Left tile: "Most developers are still using CursorAI & Chat." Subtext: 2024 tools are already outdated relative to 2025–26 agentic systems.
      • Right tile: “The Staff Engineer Curse: 10–15 years of experience.” Subtext: senior ICs lag in AI uptake while the rest of the org adopts quickly.
    • Background faintly shows a person + futuristic UI panels, stressing contrast between legacy habits and new capabilities.
    • Bottom line: slow adoption is dragging companies, generating frustration; upcoming case study: OpenAI + Codex.
    • Your note aligns: the productivity gap is enormous between adopters and holdouts.

    • We've Seen This Movie Before — historical analogies

    • Single-column layout listing past refusal patterns:
      • Watchmakers resisted electric tools → precipitated the Swiss Quartz Crisis.
      • 70s–80s drafters rejected CAD.
      • Photographers resisted digital workflows.
    • Framing question: senior engineers repeating the same resistance pattern with AI.
    • Closing line: behavior change is slow; automation always wins.

    • The New Quartz Crisis — stacked tiles

    • Top row: two side-by-side boxes.
      • Left: software ≈ mechanical watchmaking — elegant, artisanal, historically hand-crafted.
      • Right: Linus Torvalds cast as a master “watchmaker,” skeptical of vibe-coding for “real” work; echoes past artisan resistance (drafters, watchmakers, photographers).
    • Bottom wide tile: blunt prediction — old-style software development will be automated away, not over decades but 1–2 years.
    • Your note tracks: this mirrors prior industry upheavals.

    • Adoption Depends on Having UI — two-column split

    • Left column: Thesis: CLI agents are too hard; mass adoption requires a real UI. Supporting bullets:
      • Claude Code forces constant supervision.
      • You must monitor every action.
      • Users need helpers (model-driven guidance) rather than raw terminals.
    • Right column: Current agent tools aren't moving toward UI, but instead reinforcing the CLI paradigm. Rhetorical jab: has Claude Code meaningfully improved since February?
    • Comment: I like the concept of a Moldable Development environment - an UI that adapts to the task at hand

    • The Diver Problem — image-anchored metaphor

    • Dominant visual: a lone scuba diver suspended in a narrow chasm, deep underwater — evokes isolation, depth, limited mobility.
    • Text frames coding agents as divers sent “underwater” into your codebase.
      • Context window = oxygen tank.
      • A skilled diver can accomplish impressive tasks, but only until the air runs out.
    • Closing question highlights the constraint: why rely on a single diver for everything?
      • Sets up the need for multiple coordinated agents (orchestrators), not monolithic deep dives.
    • "should you send a product manager diver first right" sarcasm

  • Person/Gene Kim

    • Gene Kim Book Timeline — six-cover grid

      • Two rows of book covers arranged chronologically:
        • 2013: The Phoenix Project
        • 2016: DevOps Handbook
        • 2017: Accelerate
        • 2019: The Unicorn Project
        • 2023: Wiring the Winning Organization
        • Fall 2025: Vibe Coding (with Steve Yegge)
      • Visual cue: a decade-long arc of DevOps/engineering culture thought-leadership culminating in the forthcoming Vibe Coding era.

    • The Great Amazon API Rearchitecture — mandate slide

      • Left block: the famous Bezos API rules, with items 1–3 and 6 boxed in red for emphasis.
        • Expose all data/functionality via service interfaces.
        • All inter-team communication must go through those interfaces.
        • No other interprocess communication allowed.
        • Non-compliance → fired.
      • Items 4–5 note tech-agnosticism and requirement for externalizable interfaces.
      • Right black panel: aside noting #7 is a joke; enforcement attributed to Amazon CIO Rick Dalzell (ex–Army Ranger).
      • Sets historical precedent for industry-wide architectural pivots driven by top-down mandates.

    • Our Belief and Observations — bullet-driven thesis

      • AI's impact will reshape tech orgs at a scale ~100× larger than agile/cloud/CI/CD/mobile transformations.
      • Breakthroughs don't just adjust workflows — they reconfigure entire organizations and economies around new production modes.
      • Early case studies already hint at what this AI-driven reorganization will look like.

    • Person/Adrian Cockcroft: NoOps → NoDev — pull-quote layout

      • Medium article: Person/Adrian Cockcroft, "Directing AI Native Development" (Oct 5, 2025) — shown on the left with garden thumbnail.
      • Right excerpt:
        • Ops resisted cloud automation → "NoOps."
        • Devs will resist AI-native build systems → "NoDevs."
        • Ops eventually became DevOps; in the AI-native era, developers shift toward product-manager-style direction of AI teams rather than hand-coding.

    • Elite vs. Low Performers

    • Deployment Frequency

      • Elite: On-demand (multiple/day)
      • Low: Monthly/quarterly
      • Difference: 208×

    • Deployment Lead Time

      • Elite: <1 hour
      • Low: 1 week–1 month
      • Difference: 106×

    • Deploy Failure Rate

      • Elite: 0–15%
      • Low: 46–60%
      • Difference: 7×

    • Mean Time to Restore

      • Elite: <1 hour
      • Low: <1 day
      • Difference: 2,604×
      • Source: Google/DORA — 2019 State of DevOps Report (https://cloud.google.com/devops/state-of-devops/)

    • Our Definition: Vibe Coding

    • "When the AI writes the code, and the human supervises…"
    • "Anything besides typing in code by hand."

    • From Person/Dario Amodei's Foreword

    • Vibe coding is both inspired and misleading:
      • Inspired because it captures the experience of vaguely describing intent to an AI and watching it turn that vibe into working software.
      • Misleading because the playful name can make the practice seem trivial or unserious.
    • Core claim: vibe coding — using natural language to direct an AI model to write and iteratively improve software — is deadly serious.
    • As of mid-2025, it's the only coding game in town (according to Person/Dario Amodei)

    • Person/Erik Meijer (VB, C#, LINQ, Hack)

    • "If we are the last generation of developers to write code by hand, let's have fun doing it."
    • "Let's not waste our most precious resource — time — on things a machine can do. Time can't be stored or created."
    • Right side: Person/Gene Kim's post summarizing Meijer's talk:
      • His team's Meta-built codegen/fix models were outclassed by ChatGPT.
      • They were shocked ChatGPT could write Hack, despite the language's limited external footprint.

    • Working With Person/Steve Yegge Developer of Programming Language/Hack

      • Writing thousands of lines of code per day — described as high-quality, production-ready, well-tested.
      • Daily spend of Inline: 300–500 across 3–5 Claude Code and Sourcegraph AMP sessions, indicating heavy iterative agentic use at scale.

    • Crazy Things We've Heard

    • “As an engineer, if I’m not spending the equivalent of my salary on tokens daily, I’m doing something wrong.”
    • Implication: some orgs expect engineers to burn Inline: 500–1000/day in AI compute as normal operating cost.

    • The Value of Vibe Coding Acronym/FAAFO

    • F: Build what I want faster
    • A: Become more ambitious about what I can build
    • A: Build things alone without needing a team
    • F: Have more fun doing it
    • O: Get more options — more swings at bat, more explorations

    • Irresistible Appeal To R&D Orgs

    • Andy Glover, Director of Dev Productivity, OpenAI:
      • “One of my top goals is to get more of our developers using agentic coding, to get everyone's productivity closer to what we see in our best Codex agent users.”
      • "We're having meetings across the company about how to elevate developer productivity."

    • Booking.com (2025)

    • Person/Bruno Passos, Group Product Manager, Developer Experience
      • Presents Booking.com scale: 1.5M rooms booked daily, 500M monthly visitors, 24k employees, $23.7B revenue.
    • Person/Laura Tacho, CTO, Developer Experience
      • Shows internal metrics: Daily AI users ship 30%+ more than non-AI users.
    • Source: https://videos.itrevolution.com/watch/1121936057

    • Travelopia (2025)

    • Person/Sree Balakrishna, Technology Director, Innovation & Products
      • Describes shift from 10–12-person squads → 4–6 cross-functional engineers → future business+developer pairs.
      • States ambition: scale from $100M → $500M with the same 200 people.
      • Highlights major win: a $1.5B/year travel company rebuilt a legacy application in only six weeks using AI-accelerated development.
    • Source: https://videos.itrevolution.com/watch/1121937327

    • Fidelity (2025)

    • Person/Tapabrata Pal, VP of Architecture
      • Presents a high-level architecture: Neo4j DB → API → analytics UI, supported by repos, artifacts, containers, runtime.
      • Demonstrates an AI-assisted prototype generating a software bill-of-materials knowledge graph UI from that architecture.
      • He spent 5 days making this, instead of 6 months
      • Even leaders with their own teams are frustrated, saying, can I prove this can be done
      • Afterwards, he said, who can maintain this?
      • Enter Person/Swathi (most junior on team), outlearning everyone on org
      • Getting more headcount, because num consumers of app increase 10x
    • Source: https://videos.itrevolution.com/watch/1121938969

    • Cisco (2024)

    • Person/John Rauser, Senior Director of Software Engineering
    • Person/Anand Raghavan, Senior Director of Engineering, AI
      • Present Cisco's scale: $52B revenue, 85k employees, 94% of Fortune 100 served.
      • Describe unifying Cisco's stack using GenAI.
      • Note a major organizational mandate: 100 leaders were required to vibe-code an application within a single quarter — a top-down adoption accelerator.
    • Source: https://videos.itrevolution.com/watch/1002959532

    • AI Trust Increases With Experience

    • Scatterplot with smoothing line: months using AI tools (x-axis) vs. AI trust score (y-axis).
    • Y-axis represents how much you trust the AI.
    • Trend: trust steadily rises over the first ~20 months, then levels around ~0.55–0.6.

    • Vibe Coding Workshop For Leaders

    • Collage of apps, dashboards, data visualizations, mobile UIs, and interactive tools built during workshops.
    • These represent real artifacts produced by leaders learning vibe coding hands-on.
    • Key result: a 3-hour workshop achieved a 100% completion rate — every participant successfully built and shipped an AI-generated application.

    • Crazy Things We've Heard

    • Tech leader quote:
      • “When I told my team that I wrote an app where AI wrote 60K lines of code, and I haven’t looked at any of it…
      • …they looked at me as if they wished I were dead."

    • Crazy Things We've Heard

    • “We’ve had these stupid problems in applications that have been there for over a decade…

      
      …we’ve never worked in that codebase. We used AI to generate a fix, and we submitted a PR…
      
      …and the team accepted it."
      

    • Crazy Things We've Heard

    • “Our code velocity is so high that we’ve concluded we can only have one engineer per repo.”
    • Interpretation: once AI accelerates output, the coordination cost of having multiple humans in the same project becomes very high.