ai and ml
A lot of application engineers awareness bleak concerning the forthcoming of the profession
There's a ridiculously uncomplicated test you can use to fig out whether AI mandates at a stated business are well-considered, or are being shoved out the entrance by out-of-touch buffoons: What happens whenever an engineer documents that AI made a project worse? Are they taken seriously, or are they marginalized and eventually managed out?
Conveniently, this week Amazon published a treatise by Terk Terkowitz, a VP/Distinguished Engineer/Final Boss of Their Independent Contributor Technical Career Track, on the function of the developer in the age of AI. It was initially an inner letter sent to their application engineers and their managers rear in September, which is an engaging lag for a document that appears this confident. It additionally maps exceedingly fine to what another organizations are struggling with.
I think it's a fine document, but I can't assistance but perspective it alongside a slightly cynical eye. I see it as an attempt to create the forthcoming of the occupation of application engineering awareness small bleak. And it's emotion mighty bleak for a lot of folks correct now.
What's the context?
This treatise comes at a period of theatrical upheaval. Tech as an industry is no alien to alter (usually at 3AM, heralded by the pager), but item concerning the current AI-ification of the earth strikes many of us as basically different from former revolutions. Folks are frightened concerning their occupation prospects. "Should I provision up tech and enter farming?" is a sincere inquiry multiple group have asked me. I don't know! Tech is already a goat rodeo, so you've got a leg up if you do?
This anxiety isn't theoretical. In May, Amazon wisely close downward KiroRank, an employee-built prosperity incineration leaderboard that classified group by token usage, following the apparent item happened: group spun up junk tasks to ascend the ranks. This is now called "tokenmaxxing," and the Terkowitz treatise specifically calls it out as being at likelihood alongside "frugality." Meta did the identical item alongside "Claudenomics" and 85K workforce that identical week. And Amazon reportedly announced another 250 job cuts this week. I can't inform you that these things are related. But I can inform you that all engineer who peruse the memo assumes that they are.
What's the item say?
This document contains a lot of awesome stuff. I have a enormous flat of respect, the two individual and professional, for Amazon's Principal Engineering community; I'm led to comprehend that that regard may not be reciprocated, but that's okay. I am, by their standards, a lapsed engineer who ran distant to associate the circus as one of its lesser clowns.
Terkowitz's memo emphasizes client outcomes and engineering judgement, two things I would entirely not assign to AI any period soon. (Note: I am penning this in October of 2026, alongside the complete understanding that former comments on the capabilities anywhere AI falls abbreviated lean to age akin fine milk.) It documents activity that (today) AI tools aren't improving. That's way improved than assuming that all exception to AI inherently comes from several self-interested Luddite who's frightened of being replaced. It additionally says item astute concerning measurement: appearance at the 90th percentile of AI adoption, at those teams burning the most tokens, and inspect whether their shipment really went up commensurately. This is great! Shout it from the rooftops; we should all be doing specified things.
But still, I wonder.
The memo says “If your activity does not advantage from current AI tooling, that is a finding, not a failing.” Document it, bring it to your manager, activity through the problem. I concur alongside all term of that! The inquiry is whether anyone's promo doc is gonna last citing it.
Here's what happens next. The memo concedes that input metrics are "deeply flawed," but afterward in the next paragraph says to go onward and use them anyway, carefully. It afterward warns that non-adopters are falling behind. Those are three distinct positions inside three sentences and they're not at all aligned alongside one another.
That's not muddled; I think it's cover. A director can citation "a finding, not a failing" whenever a powerful engineer skips the AI tool, and afterward citation "falling behind" whenever it's period to PIP someone. Both are congruent alongside their stated policy; you get to decide which explanation you desire to go alongside in a stated situation.
How volition managers grip justified AI non-use? Are promotion and achievement decisions being made in a way that rewards these lengthy term outcomes? Or are they alternatively taking the uncomplicated weasel way out, optimizing for what's cheap to measure, and rewarding tool activity? Because if you provision me a metric, I volition rotate it into a target. So volition an overwhelming bulk of employees. When that happens, everyone loses — it's how KiroRank happened in the archetypal place.
Everyone's marketing tokens — including Amazon
I compose this following two days of attending GitHub Universe, anywhere they talked concerning how agents commencement approximately 40% of PRs now. And yet, Jay Parikh’s keynote disagreement was that helpful output wasn't "more code," but fairly completed outcomes: features shipped, bugs fixed, safety issues addressed, loudmouth columnists silenced, etc. The issue current is that their complete endeavor example cuts against that; they're billing in a example that charges customers on a use basis, not an output basis. There's nothing, today, that ties "AI spend" to "output that was value the effort."
Which brings us rear to Amazon. The memo says tokenmaxxing is at likelihood alongside frugality, and for Amazon’s inner disbursal line, it is. Meanwhile AWS bills Bedrock and Kiro customers by the token. Inside the building, tokenmaxxing is a issue to be stamped out. Outside it, it’s the income plan. ®
