Who builds, who decides, who manages the gap: 35 years of IT roles
Job titles follow bottlenecks. When building was slow, we hired people to manage builders. When feedback was slow, we invented agile roles. Now that code is cheap, the roles that existed to manage hand-offs are folding into fewer people.
This page is built mostly from information available on the internet, plus general industry knowledge. Dates, figures, role boundaries and company details are approximate and may be incomplete or out of date. The point is the shift in how roles evolve, not a precise record of any one organisation.
Bars show relative prominence by my judgement, not measured headcount, and the years are approximate. Tap a role name to jump to what changed. The shaded band follows the era you pick below.
Four eras, four bottlenecks
Each era had one thing that slowed software down. The roles that rose were the ones that managed that constraint. Pick an era.
Role by role: what changed, and why
For each role: what it was for, what it gained, what it lost, and why. Tap a row to open it.
Mostly gone or renamed
- Webmaster (1990s): split into front-end engineer, content designer, SEO and CMS owner.
- Release or build manager: absorbed by continuous integration and delivery pipelines.
- Systems analyst: became business analyst, then product analyst.
- Database administrator: became data engineer and cloud database operations.
- Prompt engineer (2023 boom): largely folded into AI engineer and ordinary software work as models improved.
Companies as lenses
Each organisation you listed taught the industry one thing about roles, plus a few you didn’t list that matter just as much. Tap to open.
Roles follow the bottleneck
The single idea that explains most of the chart: a role is a response to whatever currently limits delivery. When the limit moves, the role either changes or is left holding a job nobody needs.
1990s
Building is slow and expensive
Hired to manage it: project managers, business analysts, vendor managers. What broke: specs frozen too early, big-bang failures.
2000s
Feedback and integration are slow
Hired to fix it: agile coaches, product owners, test-first engineers, SREs. What broke: agile became ritual; process roles outgrew engineering practice.
2010s
Knowing what to build, and running it at scale
Hired to fix it: product managers, designers, researchers, data roles, DevOps, delivery managers. What broke: over-specialisation and hand-offs between ever more roles.
2020s
Verifying, deploying and owning results
Hired to fix it: forward deployed engineers, product engineers, AI engineers, evaluators. What may break: unreviewed output, thin skills, and no apprenticeship path.
Other forces that moved roles
- Cloud and automation (2006 on). AWS turned servers into API calls, which ended the sysadmin queue and created DevOps and platform work.
- Mobile and web scale (2007 on). The iPhone and always-on services made weekly releases normal and made design and data central.
- Public-sector failures. The UK’s National Programme for IT and the US HealthCare.gov launch pushed governments toward in-house digital teams and product thinking.
- Cheap money, then not (to 2022, then after). Low interest rates funded specialist layers and management. Rising rates and the 2022–23 layoffs flattened them.
- Build versus buy. Open source and SaaS shifted effort from writing platforms to integrating them, which feeds today’s deployment roles.
- Remote work (2020 on). Written specs and async coordination returned, which made AI drafting of documents natural.
- Frameworks that spread titles. PMBOK and PRINCE2 (1990s), Scrum (1995), SAFe (2011), the Spotify squad model (2012) and Team Topologies (2019). Each carried its own job titles into thousands of firms.
Who did the work, then and now
Work rarely disappears. It moves between people. Scrub through the years and watch each task change hands. Rows marked in orange changed owner since the previous year shown.
The whirlpool we’re in
For about 25 years roles unbundled: one job became six. Now AI is pulling them back toward a handful. Illustrative path of a single feature from idea to production:
About 1998
Project manager coordinates the whole chain.
About 2015
Delivery manager and engineering manager coordinate the team.
About 2026
Forward deployed engineers carry it into the customer’s real environment.
What’s pulling the water
- Code is cheap; knowing it’s right isn’t. The scarce skills are framing the problem, verifying output, and owning the result.
- Value is moving to deployment. Models are easy to demo and hard to fit into permissions, data and workflows. Indeed data reported by Business Insider shows forward deployed engineer postings at about 5,300 in April 2026 against about 640 a year before, from a small base. Indeed India reported a nearly 19-fold rise in June 2026.
- The hiring market is cooler overall. Indeed’s US software development postings index stood at about 77 in mid-September 2026, roughly 23% below its February 2020 level, after more than doubling that level at the 2022 peak.
- Flatter organisations. Fewer coordination layers means managers, delivery roles and BAs carry more scope, or disappear.
- The apprenticeship gap. Juniors learned by doing tasks that AI now does. Reports suggest recent software posting growth skews toward senior roles. Correlation is not proof of cause, so treat this as a risk to manage, not a settled fact.
- Accountability doesn’t compress. Someone still signs for what ships. Government adds assurance, security, procurement law and public trust, which tend to keep some coordination roles longer than in startups.
Lens for Singapore’s public sector
The arc runs from procurement and project management (IDA era), to in-house product teams after the 2016 split into GovTech and IMDA, to a question now being tested: how small can an accountable, AI-assisted product team be? GovTech’s own 2026 job schemas, in the next section, show the shift under way: engineers rewritten around reviewing AI output, managers measured on review quality rather than output volume, and policy, operations and security roles kept as full career ladders.
Inside GovTech: what the 2026 job schemas say
GovTech’s 2026 career schemas point in one clear direction: in-house engineering capability, with AI changing the engineer’s job first, specialist hand-offs folding into product teams, and coordination roles slimming down while public-sector roles hold their ground.
The direction of travel
- The engineer’s job is shifting from writing code to verifying it. Engineering ladders now expect people to direct AI tools, review what they produce, and set the standards for doing that safely. Managers are judged on quality and review, not output volume.
- Specialist hand-offs are being absorbed. Testing, DevOps and security now sit inside the engineer’s role rather than in separate teams, with a platform group providing the shared foundation.
- Coordination roles remain, but slimmer. The delivery manager role used before 2026 has been deprecated, with its people moved into a newer role. Business analysis and project management have short career paths written in product and agile terms. Senior coordination runs through programme management, and vendors are still part of delivery.
- Public-sector roles hold their ground. Policy translation, product operations, security governance and agency technology leadership keep full career paths. This is the work AI is least likely to compress.
- AI adoption is uneven. Engineering and leadership have written AI into expectations. Product and programme roles mostly haven’t yet, so the product-manager and product-engineer overlap seen in industry hasn’t reached these ladders.
Limits of this view
Based on 16 of 28 role schemas; data and AI roles weren’t available. Schemas describe expectations, not team structure or headcount. Roles discussed elsewhere on this page, such as QA, product engineer and forward deployed engineer, have no GovTech family here but may exist under other titles. The history of the shift, from the IDA era to today’s product teams, isn’t covered.
What to take away
- Roles are responses, not identities.Ask what bottleneck a role was hired to manage. If that bottleneck has moved, expect the role to change.
- Work moves; it doesn’t vanish.Testing, operations and requirements didn’t disappear. They moved into developers, platforms and product teams.
- Titles lag practice.Job boards like Indeed show a title after the work has already changed, so use them to confirm trends, not to spot them early.
- Big tech writes the playbook; others copy it years later.Copying titles without the engineering discipline behind them is how agile became ritual.
- Each wave unbundles, then a later one rebundles.Expect today’s product engineer to fragment again once its limits show.
- Durable skills.Problem framing, domain knowledge, verification, communication and owning outcomes have survived every shift on this chart.
Where this could be wrong
Dates are approximate and the bars are my judgement. Company histories are from public sources and my own knowledge. GovTech details now draw on 16 of its 28 role schemas, so its data roles, team structures and history are still unverified. Hiring statistics are reported by third parties and are small-base or US-centric in places.
Sources worth opening
- Agile Manifesto
- Google SRE book, introduction
- UK Government Digital Service
- Indeed software development postings index (FRED)
- Indeed Hiring Lab
- GovTech career schemas (Appraise), 16 of 28 role families, versions dated 30 March to 19 October 2026. Internal documents supplied for this page.
- Forward deployed engineer postings in India, Indeed report coverage
Built as a teaching aid. Last reviewed October 2026.