Key findings
- 16 of the 18 sources that discuss the spec workflow are positive, and 8 say planning first improved the result.
- All 12 sources that discuss review burden raise it as a cost, from over-built code to an agent that loops and misses the root cause.
- Engineers split 7 for and 6 against on whether specs pay off. On whether Kiro becomes their main tool, 6 made it so and 7 kept another tool.
- Only 11 sources speak for a team, a leader, or an organization. Seven are positive, three come from Amazon or AWS engineers, and one is an AWS Organizations administrator who could not assign paid plans.
- 50 accounts apply as written and 19 need re-checking, most of them launch-period model complaints. Four are out of date: 3 rest on request pricing Kiro retired in September 2025, and 1 on a memory workaround that Kiro's native Memory replaced in October 2026.
Engineers praise Kiro's spec workflow most and name the review load it creates as its main cost. Most of what they have reported since launch still applies to Kiro as sold today. GILD read 73 public accounts published between July 14, 2025 and October 2, 2026, by named engineers who include a Thoughtworks distinguished engineer and an engineering manager. The set also includes Johann Rehberger's independent security test.
Who wrote the accounts
The 73 sources are 53 Hacker News comments, 4 GitHub issues, 1 independent security test, and 15 articles. The articles are 8 DEV Community posts by named authors, one martinfowler.com article, one Substack post, one personal blog, three blog posts from AWS consultancies, and one conference interview.
Individual engineers wrote 62 of them. The other eleven describe two very small teams, one engineering manager, four technical leaders at AWS consultancies, and four organization-wide uses. Six Amazon or AWS engineers wrote seven of the Hacker News comments, and some describe required Kiro use from before Amazon offered Claude Code and Codex internally.
What still applies
Kiro opened early access on July 14, 2025 with Claude Sonnet 4 and 3.7, a waitlist, and daily limits. Fifteen accounts complain about performance in that period, so check them against the current models. Two more describe bugs in older builds: a launch-day prompt injection AWS fixed in v0.1.42 and a January 2026 context loop. Two Caylent leaders praised its spec mode in launch week. Three depend on per-request vibe and spec pricing, which a credit pool replaced on September 15, 2025. GILD marks those three Out of date, along with one memory workaround that Kiro's native Memory replaced in October 2026.
Thirty-two earlier accounts still apply because specs, steering, and hooks behave the same way today. Another 18 describe the current product, and 17 came out after Kiro Web launched on May 18, 2026.
What teams do differently
Inside AWS, where most engineers use Kiro, teams share AGENTS.md files and team-specific skills, including on-call skills. One organization built an on-call MCP server. A platform engineering manager used one spec and parallel CLI subagents to migrate Terraform locking across 30+ repositories in under 90 minutes, against a manual estimate of 4 to 6 hours. One company enables overages for all engineers and adjusts tiers by usage.
AWS reports that six senior engineers rebuilt part of the Amazon Bedrock inference engine in 76 days with Kiro as their main environment. AWS wrote that case study itself, so GILD lists it apart from the 73 public accounts.
- Keep steering, hooks, and skills in the repository so every member gets them.
- Decide in advance which changes get a spec and which go straight to chat.
- Size each agent task to one reviewable commit.
Where individual engineers struggle
Eleven of 12 sources discussing reliability raise concerns. They report early model errors, account suspensions in March 2026, and MCP servers that broke after a CLI update. An editor crashed 3 times in one day in August 2026. Spec design generation failed three times in October 2026 while still using credits. Eight describe spec overhead, such as a small bug that grew into 4 user stories with 16 acceptance criteria, task lists too long to review, and agents that ignored the spec. Five describe the agent losing or exhausting its context.
What to test in a pilot
Engineers split on whether specs repay their cost, so run the same tasks with and without specs and count accepted changes and review time. Set a spec size limit and a rule for bug fixes before the pilot starts. Track credits per accepted change for each model. Put pilot engineers on managed sign-ins so AWS does not use their content for service improvement.
Methodology
GILD opened every source on October 12, 2026. It recorded the author, publisher, publication date, overall sentiment, kind of evidence, and positions taken, in the same record format GILD uses for Claude Code. A verbatim excerpt records each author's role. GILD read Hacker News comments through the Hacker News API, DEV posts through the DEV API, and GitHub issues through the GitHub API. GILD added Johann Rehberger's security test on October 25, 2026 as an independent evaluation. Only English-language sources by a named author or a public Hacker News handle count, so DEV posts by unnamed authors are out. GILD also excluded AWS employees writing as the vendor, news write-ups, hackathon promotions, and setup tutorials with no reported experience. Amazon engineers describing their own use stay in, labeled as Amazon. GILD reports a finding only when at least 5 sources from 3 different authors discuss it. Engineers who post in public choose to, so the counts describe what those 73 accounts report about Kiro. Confirm fit with a pilot spec on your own code.
Products in this piece
Sources
- Official documentation
- 5
- Public source
- 12
- Kiro docs: Specs
- Kiro case study: Rehrig Pacific (Oct 2, 2026)
- Kiro blog: Announcing new pricing plans and Auto, our new agent (Sep 15, 2025)
- Kiro pricing
- Kiro docs: Cloud sessions
- Birgitta Böckeler, Thoughtworks, on martinfowler.com: Understanding Spec-Driven-Development: Kiro, spec-kit, and Tessl (Oct 15, 2025)
- Yehuda Cohen: Developing with Kiro: Amazon's New Agentic IDE (Jul 14, 2025)
- Geshan Manandhar: Using Spec Driven Development with AWS Kiro to add the last updated date on Eleventy blog (May 3, 2026)
- Hacker News comment by sparrc, an AWS engineer, on Kiro use inside AWS (Jun 9, 2026)
- Davide de Paolis: Kiro showcase: Automating Changes Across Several Repos with Spec-Driven Development and Custom Sub-Agents (Apr 7, 2026)
- Francois Dexemple: Brilliant, Broken, and Frustrating: My Deep Dive into Amazon's Kiro AI IDE (Oct 24, 2025)
- Hacker News comment by thegagne on 8 hours and 200+ credits of spec work (Jan 8, 2026)
- Hacker News comment by wenc on using Kiro IDE as a spec generator (Mar 11, 2026)
- Hacker News comment by iLoveOncall on required Kiro use inside Amazon (May 5, 2026)
- Hacker News comment by huey77 on a suspended Power plan account (Mar 22, 2026)
- Hacker News comment by forgotaccount3 on company-wide overages and tier shifts (May 28, 2026)
- Akira Tateishi: When a Kiro CLI Update Silently Breaks Your MCP Servers (May 21, 2026)