Methodology
How eras, scores, and sources work
Technology Rewired is my editorial read of how each category of software is changing. Every page follows the same template so categories can be compared side by side, and every factual claim links to a source.
What are the five eras?
- Era 1: On-premBefore 2005
- On-prem: every function wired by hand, in its own closet. Software ran in-house, and people moved the work between systems.
- Era 2: SaaS and cloud2005 to 2020
- SaaS and cloud: everything plugs into one backbone. Per-seat pricing, browser apps, and APIs connect the functions.
- Era 3: AI-assisted2020 to 2024
- AI-assisted: a copilot on every chip. Software suggests, scores, and drafts. A human still drives every step.
- Era 4: Agentic2024 onward
- Agentic: agents carry work across functions on their own. Support hands to sales, sales to finance, and people approve the exceptions.
- Era 5: Next 5 years2026 to 2031
- Next: the wiring fades into the background. Most routine work runs ambient and autonomous, and humans set intent and approve.
Era boundaries are typical, not fixed. Each page states its own years, because identity went to the cloud later than email and support met agents earlier than finance.
How is the autonomy score calculated?
Three sub-scores, each 0 to 5 with one decimal. The overall score is their mean, rounded to one decimal, and the build fails if it is not. The agentic-era score always equals the overall score, and the next-era score always equals the five-year projection.
| Score | Tasks automated | Approval load (higher = fewer) | Production maturity |
|---|---|---|---|
| 0 | No step done by software | A human does every step | Research demos only |
| 1 | Suggestions only | A human approves every action | Pilots and betas |
| 2 | Agent completes a narrow set of steps | A human approves most actions | GA from startups, early customers |
| 3 | Agent completes most routine steps for a meaningful share of volume | A human approves only consequential actions | GA from several vendors with named customer deployments |
| 4 | Agent resolves the majority of volume end to end | Exception-only review | Mainstream among category leaders |
| 5 | Virtually all work | No routine approvals | The default way the work is done |
Category scores on the board are the mean of their deep dives. Categories with no deep dive yet carry a category-level estimate, labelled as backlog.
How are claims sourced and verified?
- Primary sources first: vendor announcements and documentation, regulatory filings, standards bodies. News coverage is used when no primary source exists.
- Vendor capability lines are checked against the vendor's own site. Announced but not generally available is written as announced.
- A claim that could not be confirmed is marked "being verified" on the page rather than guessed.
- Vendor-reported performance numbers are attributed to the vendor, never stated as fact.
- Every era block carries its own last-verified date. A page not verified for 120 days is flagged as needing a refresh, and the build refuses to ship it otherwise.
How is Rewired Metrics counted?
The count covers 750 final IPO prospectuses (424B4 and 424B1) of US-listed tech companies filed from 2004 to 2026. The headline cohort is software companies with a US business address. A prospectus counts once per metric. A match is not always a disclosure, so a reviewed sample of each metric splits matches into the company reporting the metric about itself and passing mentions, and the reported share is estimated from that review.
- Wordings per metric are fixed in a published list. Each prospectus counts once per metric.
- A sample of up to 20 matches per metric per era is opened by hand and marked reported, mention or false. Wordings under 80% precision are tightened and the count is run again.
- Measured precision per metric: Annual recurring revenue (ARR) 100%; Net revenue retention (NRR) 100%; Gross revenue retention 96%; Customers over $100,000 100%; CAC payback, magic number and LTV/CAC 100%; Daily and monthly active users (DAU, MAU) 100%; Billings 100%; Remaining performance obligations (RPO and cRPO) 100%; Adjusted EBITDA 100%; Free cash flow and free cash flow margin 100%; Gross margin 100%; Stock-based compensation 100%; Consumption and usage-based revenue 100%; Revenue per employee 100%; Artificial intelligence mentions 100%; AI agents and outcome-based pricing 100%.
- Years with fewer than 15 prospectuses are labelled small samples. The count shows what companies chose to lead with at IPO; metrics that live in earnings releases and proxies, such as the Rule of 40, need other forms.
How do refreshes and predictions work?
Every six to eight weeks I re-check the agentic era, the prediction, vendors, scores, and links. The past eras rarely change. Score history is kept, and any move of half a point or more gets a line in the changelog. Predictions are never rewritten: they are marked on track, early, or off track, with the signal that moved them. I keep the same discipline on my public predictions scoreboard.
Further reading
Questions
- What does an autonomy score of 3 mean?
- Agents complete most routine steps of the job for a meaningful share of the volume, a human approves only the consequential actions, and several vendors ship the capability to named customers.
- How often are Technology Rewired pages updated?
- Every six to eight weeks. Each refresh re-verifies the agentic era, the five-year prediction, vendors, scores, and internal links. Pages not verified for 120 days are marked as needing a refresh.
- Are vendors paying to be listed?
- No. Vendors are chosen editorially and their capability lines are checked against their own websites. Being listed is not an endorsement.