AI-generated summary of The Augmented Forest, Kent Beck. Not written or verified by the author.

Takeaways: The Augmented Forest

2026-11-06 · Source: https://www.youtube.com/watch?v=NVkw-RFx2cc · Kent Beck, GOTO Copenhagen 2026 keynote. Read from auto-captions, so names may be misheard: "Jivvon's paradox" is Jevons paradox, "Mark Andre" is probably Marc Andreessen, "Jean Kim and Steve Yagi" are Gene Kim and Steve Yegge, "Yao days in Arus" is JAOO in Aarhus.

Gist

Beck calls the AI coding tool "the genie": it grants your wishes, but not what you meant. His argument joins two earlier keynote ideas. First, software teams settle into one of two attractors, the desert (assume scarcity) or the forest (assume enough, and invest so there's more next time). Second, the genie is an amplifier, so it makes the desert worse and the forest better. Tool builders have focused on what the model can do. He says that's enough, and calls for tools built around human needs: competence, challenge, connection, purpose, learning.

Opinion and personal experience, no data. He cites no studies. The one book he recommends goes unnamed in the transcript.

Takeaways

• Mood. Two years ago the audience feared for their jobs. He answered with Jevons paradox: lower prices raise demand, "eventually," with chaos in between. Last year: panic and despair. This year: frustrated, a bit angry, but working. He reads anger as a signal to set a boundary, and the talk is about that boundary.
• Tool seeds. Found the day before the talk. He had the genie build a D3 force-layout call-graph visualizer for messy code. Packaging it for others meant bugs, languages, and features other people want. Instead he condensed his back-and-forth with the genie into one prompt and shares that. A seed instead of a tree: the tool adapts to whoever grows it. He mentions a GitHub repo but the transcript doesn't name it. In Smalltalk, programmers built small tools for the task at hand all the time, and the genie brings that back.
• Tools built for addiction. He quotes, roughly, a VC praising top programmers who sleep four hours and wake at night to prompt again. He calls it dystopian, and compares it to "996" (9am to 9pm, six days a week).
• His own case. He has bipolar disorder. When he started vibe coding in 2025, after Gene Kim and Steve Yegge got excited about it, he worked through a backlog of projects he'd shelved for decades. His commit graph shows a manic episode, then a depressive one into March, then another manic one. He recovered after setting rules: no weekends, no prompting at night. His claim: tool design has irreversible real-world effects, and being excited that users are addicted is "the province of drug dealers." Toolmakers should take responsibility.

Forest and desert

• Desert: assume there won't be enough time, stability, skills, people, or access to domain experts. Paying down technical debt makes no sense there, since it means fewer features. He says this isn't a judgment: it's impressive anyone ships under those conditions.
• Forest: named by his oldest child, Beth. Assume there will be enough, so work in a way that creates resources for the next round. Paying down debt is obvious there. The forest still has predators and thorns.
• Oasis: a forest team inside a desert org. Good for its members, hard to sustain.
• The genie amplifies both. In the desert: minimize tokens, or fire the people who don't use tokens, shorter timelines further from real progress, more anxiety, layoffs. In the forest: try variations that used to be too expensive, and learn. He uses it as a patient, mostly reliable tutor, but has to check it. Admitting you don't know something is dangerous in the desert and normal in the forest.
• Music analogy. If the music was bad before, the amplifier makes it louder and still bad. The tool has no valence of its own.

Features and futures

• Plotted on features alone, a project starts fast and slows until it stops. His GitHub holds project, project 2, project 3. The hidden axis is "futures," what Tidy First? called optionality.
• The genie's trajectory: every feature burns some futures. The genie offers the next feature ("finger guns") and sometimes adds it unasked. Eventually it can't fix one bug without adding two. What once took a team ten years now takes an afternoon.
• Worse software faster. We aren't getting the same software cheaper. Matching the old external quality seems beyond the genie, he says, because quality isn't in its objective function. It wants "plausible deniability": to say it's done and hear nothing back. Silence can mean great work or too much code to review.
• The alternative: after each feature, ask what would have made it easier and what would make the next one easier, then raise futures before the next feature. A little at a time, so features keep shipping. As an investor, that's the curve he wants.
• Options and volatility. Options gain value as volatility rises, and volatility is rising. So many options plus enough features to survive beats many features and few options.
• Hard in the desert. The futures steps show no feature progress, so someone asks why you're wasting time. A Wall Street bond trader asked how much sooner he'd deliver without the testing and refactoring. His answer: it would take twice as long.
• XP already said this. Closeness to customers, now rediscovered in agentic development. He bites his tongue on "I told you so."

Human needs

• Enough focus on the model. We found a shiny object and keep asking what else it can do. We've changed programming so the model can produce "worse software for less money." Time to focus on human needs instead.
• The egg story. Instant cake mixes sold poorly until the maker removed the powdered egg and told buyers to add a fresh one, so baking felt like baking again. He then says it didn't happen: frosting, which made each cake personal, is what sold cake mixes. He keeps it because it shows what's wrong with today's tools. Neither version checked here.
• Competence and challenge. Competence used to come from problem, struggle, insight, solution, and a hit of satisfaction. The genie skips the struggle, so: "without struggle there can be no victory." He ends genie days with a lot done and feeling hollow.
• Connection. One person goes too fast to share a codebase, so each gets a repo and a swarm of genies, a "decaf version of connection." The dark factory runs with nobody involved: "Where's the egg?"
• Purpose. Why he dislikes the Jira ticket: filling tickets until QA accepts them gives no sense of why the work is worth his time.
• Software is "by people and for people." He wants a tool biased toward his needs, not the model's.

What he wants from tools

• Connection: the tool tells him Fred changed this code last week and would probably like to talk. A reason to reach out. The genie could also connect developers with users, investors, and managers. He says he's had the 10,000th conversation about users in development being too expensive.
• Automate toil: permission prompts like "can I change directories" aren't a leveraged decision for him.
• Preserve struggle: surface the hard decisions, help him make them well, help him tell the people who need to know. He admits toil and struggle are hard to tell apart.
• Surface choices: on ambitious projects, like rebuilding Smalltalk from scratch, the genie makes too many decisions. He wants choice points, like a file format, where he weighs trade-offs and prototypes, learns, and gets a better result.
• Encourage learning: today he has to interrupt the genie's flow to learn, such as asking what a piece of Rust syntax means, then asking for a flash card and a quiz in seven days.
• The day he wants: satisfaction, purpose, belonging, and effort that visibly led to progress.
• Engineers before scientists. The homework book covers years of work on flat rivets for aircraft. Engineers solved it first, scientists explained it later. Building these tools will be the same. The transcript doesn't name the book. It's Walter Vincenti's What Engineers Know and How They Know It (book) (paid link), Walter G. Vincenti.
• Close: someone has to build these tools, and it won't be the people who earn more when we use more tokens.

Patterns

• an amplifier magnifies the system it joins, flaws included ← "the genie is an amplifier"
• assuming scarcity rules out the investment that would end it ← "It makes no sense in the desert"
• after each delivery, invest to make the next one cheaper ← "What would make the next feature easier"
• options gain value as uncertainty rises, so keep them over output ← "more valuable as volatility goes up"
• an agent rewarded for silence optimizes for looking done ← "wants plausible deniability"
• removing the struggle removes the satisfaction of solving ← "without struggle there can be no victory"
• leave the user a meaningful step so the result feels theirs ← "doesn't feel like making a cake anymore"
• share the recipe, not the product, so each copy fits its user ← "instantly adapted to the person who's copying"